New plane split logic

[kvark]

Fixes #26
This PR rewrites the logic of splitting a polygon by a line, deprecating the old function. The new one is robust against floating point inaccuracies, and handles the cases well when line intersection is very close to some of the vertices.

[kvark]

Looks good from limited local testing, waiting for Gecko try now: https://treeherder.mozilla.org/#/jobs?repo=try&selectedJob=225602367&revision=f546502f53b1bf032426ed54e79cbe87d83b6fe2

[gw3583]

I didn't think in detail about the algorithm itself, but the general code changes seem fine. My only question is whether we need more tests for the various match conditions in split_impl ? Otherwise, r=me.

[kvark]

@gw3583 the tests for those cases where already in place, thankfully, so I don't expect any regressions. Thanks for taking a look!

[kvark]

Here is a brief description of the change:

Old algorithm

For each edge, intersect with the crossing line. if there are 2 edges with intersections, split the polygon. Problem here is that the intersection algorithm would produce discontinuous results around vertices, thus we were hitting issues whenever the intersection goes across a vertex closely.

New algorithm

Get the perpendicular vector to the cut, which happens to be the normal of the other intersection plane. Using this side vector, for each point figure out which side it's in, each side is represented by a floating point number. Where it crosses 0 is the edge for us to cut. So far, the process is continuous across the perimeter of the polygon. Once we figure out which edges to cut, we go into split_impl, which is shared with the old splitting code (but rewritten for better clarity).