Fix rounding error between triangle area calculation and intersection calculation

[mfogel]

Currently, both the triangle area calculation in signed_area.js and the intersection calculation in segment_intersection.js are used to determine if two line segments are collinear. A signed area of 0 indicates collinearity, while a intersection calculation of 2 also indicates collinearity.

Segment intersection is used here while signed area is used here and here.

It's possible (due to rounding errors) for the signed area calculation to return a non-zero result (indicating the segments are not collinear) while for the same two line segments, the intersection calculation will calculate two intersections (indicating the segments are collinear). When this happens, segments can get inserted into the sweep line but never removed - messing up order of segments in the sweep line which in turn messes up the calculation of whether other line segments should be included in the final result or not.

This PR adds a test case demonstrating that sutiation, and implements a fix.

[rowanwins]

I've had a quick play and this pull request in conjunction with pull #58 appears to resolve issue #35 .

The only thing that still seems to be broken on it is Difference B - A (see the http://localhost:5000/demo/#overlap_two), this should return an empty array I think, but it's looping to infinity and beyond!

Great work @mfogel

[rowanwins]

So i just replaced avl with splaytree and this appears to resolve the issue above with Difference B - A

[w8r]

Cooool, how are the benchmarks? Splaytree is slightly slower on insertion