It works a little bit like a long division with two basic differences:

1. Instead of one digit at a time, you handle "slices" of two digits at a time (the pair of digits to the left of the decimal point being one such slice).
2. You get each new digit by trying at each step to "extend" twice the previously found root with another digit to that the product of this extension by the (largest possible) new digit does not exceed the previously computed remainder.

1. Remainder is 2, Root is 0, DoubleRoot is 0.
   The largest digit "?" which is such that "0?" times "?" is no greater than 2 happens to be 1 (save this as the first digit in the square root). "01" times "1" is 1, which I subract from 2, the result is 1, which gets augmented by the "next slice" of unprocessed digits, namley "00". So, the new remainder is 100.
2. Remainder is 100, Root is 1, DoubleRoot is 2.
   What's the largest digit such that "2?" times "?" does not exceed 100. Well it's 4 (save this as the second digit in the root). "24" times "4" is 96, which I subtract from 100 and augment with the next slice of "00" digits to get 400 as the new remainder.
3. Remainder is 400, Root is 1.4, DoubleRoot is 28.
   The largest digit such that "28?" times "?" does not exceed 400 is 1 (which is thus our next digit). The new remainder is 400-281=119 augmented by the next "00"; 11900 is the new remainder.
4. Remainder is 11900, Root is 1.41, DoubleRoot is 282.
   The next digit is 4 for which "282?" times "?" is 11296. 11900-11296 is 604; new remainder is thus 60400.
5. Remainder is 60400, Root is 1.414, DoubleRoot is 2828.
   The next digit is 2 for which "2828?" times "?" is 56564. 60400-56564 is 3836; new remainder is thus 383600.
6. Remainder is 383600, Root is 1.4142, DoubleRoot is 28284.
   The next digit is 1 for which "28284?" times "?" is 282841. 383600-282841 is 100759; new remainder is thus 10075900.
7. Remainder is 10075900, Root is 1.41421, DoubleRoot is 282842.
   The next digit is 3 ... etc.

The square root of 2 is: 1.414213562373095048801688724209698...

No.  Just consider the sequence whose general term is (P,P²): (1,1) (2,4) (3,9) (4,16) ... which does not even contain 3 collinear points. (These are points on a parabola, but points on any infinite convex curve would do!)

No, there need not be an infinite subset of collinear points in this case either...  Consider the sequence of points whose general term is (floor(Ön),n): (1,1) (1,2) (1,3) (2,4) (2,5) (2,6) (2,7) (2,8) (3,9) (3,10) ... The distance between consecutive points is indeed bounded; it is at most equal to Ö2.  No more than 3 points of this sequence may belong to a line that's not vertical, while only finitely many may belong to a vertical line (namely, 2p+1 points at abscissa p).

This does not settle the more interesting question of knowing whether there exist for any given M (say M=1000000) at least M collinear points in such a sequence... (See next article.)

References:   [Affirmative answer to the second part of the question.]
This obscure result is indeed sometimes known as the Gerver-Ramsey theorem:  Joseph L. Gerver and L. Thomas Ramsey, On certain sequences of lattice points, Pacific J. Math. 83 (1979) 357-363, MR 81c:10039. 
    It is, in fact, enough to assume only that the gaps [the distances between pairs of consecutive points] are bounded on average (which is to say that the sum of the first n gaps is less than nB, for some positive constant B):  Carl Pomerance, Collinear subsets of lattice point sequences--an analog of Szemerédi's theorem, JCT-A (1980) 140-149, MR 81m:10104.

This is true for any integer M.  Here's my proof, which may or may not be the "classic" one which Marc Ordower will not share... 

We may assume that: 

• The range of the sequence is infinite.  (Or else the result is trivial, since at least one point must appear infinitely many times, so that infinitely many elements of the sequence are collinear because they are equal). 
• Consecutive points are distinct.  (The theorem so restricted is clearly equivalent to the more general one).

Without loss of generality, we may also assume that consecutive points are adjacent (that is, they are one unit of distance apart).  If the result holds in this special case, it holds for the general case because of the following argument:

For any sequence S of points (X(n),Y(n)) where two consecutive points are at most D units apart, we may consider the sequence C obtained by removing from the sequence (floor(X(n)/D),floor(Y(n)/D)) any consecutive elements which happen to be equal.  Since C is a sequence of adjacent points, the restricted theorem implies that, for any M, there is at least one straight line (with some rational slope q) going through MD² elements of C.  Well, C clearly corresponds to D by D square cells which contain each at least one element from the original sequence S.  To the above line of slope q corresponds a set of at most D² lines of slope q which contain each at least one integral point from each of those MD² aligned cell.  The pigeonhole principle then tells us that at least one such line contains at least M points from the original sequence S.  In other words, the theorem for adjacent consecutive points does imply the general case where consecutive points are only known to be less than some distance D apart.

We now complete the proof by ...

Ford circles are named after Lester R. Ford, Sr. (1886-1975) who introduced the concept in 1938  (American Mathematical Monthly, volume 45, number 9, pages 586-601).  The famous Ford-Fulkerson optimisation algorithm (1956) is named after his son, Lester Randolph Ford, Jr.  (who was born in 1927).

The Ford circle of the rational number  p/q  (expressed in lowest terms) is the circle of  diameter  1/q²  tangent from above to the x-axis at point p/q.

Ford circles of irrational numbers are of radius 0 and reduce to a single point.  The line  y = 1  is sometimes considered to be a degenerate Ford circle of infinite radius associated with  ¥  (unsigned infinity).

Remarkably, two Ford circles never overlap.  The Ford circles associated with two irreducible fractions  a/b  and  c/d  are  tangent  to each other if, and only if, ad  and  bc  are consecutive integers.  The largest Ford circle between such tangent Ford circles is associated with the  mediant  fraction  (a+c) / (b+d).

The  n^th  Farey series  is the increasing sequence of all the rational numbers  (irreducible fractions)  between 0 and 1 whose denominators are n or less.  For example, the sixth  Farey series  is:

0, 1/6, 1/5, 1/4, 1/3, 2/5, 1/2, 3/5, 2/3, 3/4, 4/5, 5/6, 1

Each term of a Farey series is the mediant of the two terms that surround it  (the three corresponding Ford circles are thus pairwise tangent).

In 1816, the geologist John Farey, Sen. (1766-1826) published this remark in a short letter to the  Philosophical Magazine, entitled "On a curious Property of vulgar Fractions".  This attracted the attention of Augustin Cauchy, who supplied a proof in  Exercises de mathématiques (1816).  Cauchy is responsible for the enduring attribution of these finite sequences to Farey.  In 1802 (14 years before Farey and Cauchy) one C. Haros had studied the 99^th Farey series, in a paper about the approximation of decimal fractions by common fractions.

Let's label each node of a binary tree with a triplet of integers so that the left and right offsprings of a node labeled (x,y,z) are respectively labeled (x,x+y,y) and (y,y+z,z).  The label of the root determines the labels assigned to all the nodes.

Let's call   numerator tree  the tree so labeled whose root bears the label (0,1,1).  Call  denominator tree  the one whose root is labeled (1,1,0).

We use a graphical representation, where each offspring is assigned a box half the size of its parent's.  In each box, only the middle integer of the "label" is shown.  The other two values are simply found at the top of each bounding vertical line.  (In the middle of an ascendant's box, except for a leftmost or rightmost node.)

The Stern-Brocot tree is the binary tree whose nodes are labeled with the rational number p/q, where p (resp. q) is the middle integer in the label of the corresponding node of the above numerator tree  (resp. denominator tree).  Remarkably, all rational numbers are thus obtained  in irreducible form.

The rational numbers which appear in the left half of the tree are between 0 and 1.  A Farey series is obtained when the numbers in the first half of a line are interspersed with the numbers sitting at the top of the box boundaries.  For example, the fourth Farey series is obtained from the last line shown above:

0, 1/4, 1/3, 2/5, 1/2, 3/5, 2/3, 3/4, 1

Any positive rational appears once and only once in the Stern-Brocot tree, at a location determined by its  continued fraction expansion (CFE)...

To reach the positive rational whose CFE is  [q₀,q₁,q₂...q_n]  go q₀ steps right, q₁ steps left, q₂ steps right, and so forth alternately...  then  go back one step  (either that, or perform only q_n-1 steps in the last stage).

The Stern-Brocot tree was discovered (at least) twice:

• In 1858 by Moritz Stern  (Über eine zahlentheoretische Funktion. Crelle's Journal  55:193-220).  Moritz Abraham Stern (1807–1894) is credited with noticing the genius of his student Bernhard Riemann (1826-1866) early on at Göttingen.  Stern was also one of the main teachers of Richard Dedekind (1831-1916) who was Gauss's last student (receiving his doctorate in 1852).  In 1858, Stern was appointed to succeed Gauss (1777-1855) as professor ordinarius of mathematics at Göttingen University.  He was the first jewish Ordinarius in a German university. 
• In 1861, by Achille Brocot  (Calcul des rouages par approximation, nouvelle méthode. Revue Chronométrique  3:186-194).  Achille Brocot (1817-1878)  was a famous innovative clockmaker, who stumbled upon the Stern-Brocot tree while computing optimal gear ratios.

0,1,1,2,1,3,2,3,1,4,3,5,2,5,3,4,1,5,4,7,3,8,5,7,2,7,5,8,3,7,4,5,1,6,5,9,4...

In 1982, Edsger Wybe Dijkstra (1930-2002) called the sequence of Stern numbers "fusc" because it possesses a curious obfuscated property:  If the sum of two indices, n and m, is a power of 2, then b(n) and b(m) are  coprime.

The sequence of the ratios of two consecutive Stern numbers  u_n = b(n) / b(n+1)  runs through all nonnegative rational numbers (in reduced form) just once! 

0, 1,1/2, 2, 1/3, 3/2, 2/3, 3,1/4, 4/3, 3/5, 5/2, 2/5, 5/3, 3/4, 4, ...

Moshe Newman  is credited with the discovery of a very nice alternate definition of this rational sequence, namely:   u_o = 0   and   u_n+1 =  f (u_n )   where

f (x)  =  1 / (1 + 2 ë x û - x )

When this sequence is displayed as a binary tree,  each level  features the same fractions as in the Stern-Brocot tree, but at different locations:  The respective positions [0 = leftmost] of a given fraction are mirror images in binary numeration.

In this tree, the left child of   a / b   is   a / (a+b) ,  the right child is   (a+b) / b.  Thus, if the parent fraction is in  lowest terms, so are both offsprings...

References and History :

• Gotthold M. Eisenstein (1823-1852)
  "Eine neue Gattung zahlentheoretischer Funktionen [...]" (1850)
  Verhandlungen der Königlich Preussischen Akademie der Wissenschaften zu Berlin 
• Moritz A. Stern (1807-1894)
  "Über eine zahlentheoretische Funktion" (1858)
  Journal für die reine und angewandte Mathematik,  55:193-220. 
• C. Giuli  and  R. Giuli
  "A primer on Stern's Diatomic Sequence" (1979; some errors)
  Fibonacci Quarterly,  17:2 (1979) 103-108, 246-248, 318-320. 
• Neil J. Calkin   and   Herbert S. Wilf
  "Recounting the Rationals" [ pdf,  July 6, 1999 ]
  The American Mathematical Monthly,  107 (April 2000) 360-363. 
• Igor Urbiha
  "Tree of fractions and Dijkstra's fusc function" (March 27, 2002)
  Seminar on Number Theory and Algebra, University of Zagreb. 
• Jeremy Gibbons, David Lester and Richard Bird
  "Enumerating the Rationals" [ pdf ]

This result is due to Georg Pick (1852-1949) who published it in 1899  ("Geometrisches zur Zahlenlehre"  Sitzungber. Lotos, Naturwissen Zeitschrift  Prague, Volume 19, pages 311-319).  Pick was instrumental in the appointment of Albert Einstein to the German University of Prague and became a close friend of Einstein's during his stay in Prague, from 1911 to 1913.

Pick's theorem became famous in 1969, when it appeared in the third edition of the popular book  Mathematical Snapshots  which Hugo Steinhaus (1887-1972) had first published in 1938...

Lattice points  (which Pick called  reticular points)  are points whose coordinates are integers.  A lattice polygon is a polygon whose vertices are lattice points.  In the plane, a lattice polygon may be dissected into lattice triangles which have no lattice points (besides their vertices) inside of them or on their borders.  Therefore, Pick's formula for the area of a lattice polygon is a consequence of the following (surprising) special case:

A lattice triangle which has no lattice points besides its vertices
(i.e., inside of it, or on its edges)  has a surface area of  ½