Add polygon test showing bad tessellation result + fix

[kintel]

I believe this shows that a relatively complex "branchy" polygon with a single triangular hole causes Triangulate() to emit some pretty funky triangles.

This polygons is a very slimmed down version of the polygon emitted by https://github.com/openscad/openscad/blob/master/examples/Advanced/module_recursion.scad

The polygon itself is a result or a union of a large number of 2D rectangles using Clipper V1.
It's very possible that there is some precision-related issue causing this, but I believe that a robust triangulator should be able to handle this kind of abuse.

PS! validated by emitting the triangles as an OBJ file and rendering that externally:

std::cout << "# polygon-test.cc\n";
  for (auto poly : polys) {
    for (auto v : poly) {
      std::cout << "v " << v[0] << " " << v[1] << " 0\n";
    }
  }
  for (auto tri : triangles) {
    std::cout << "f " << tri[0] + 1 << " " << tri[1] + 1 << " " << tri[2] + 1
              << "\n";
  }

[elalish]

Indeed, this is a valid polygon and a good test case. Turns out a couple of valid ears were being rejected (positive cost) due to verts near their tip. I've improved their neighbor checking to get correct costs, so now everything proceeds properly.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.23%. Comparing base (d437097) to head (0e6d0db).
Report is 32 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #824      +/-   ##
==========================================
- Coverage   91.84%   89.23%   -2.61%     
==========================================
  Files          37       61      +24     
  Lines        4976     8482    +3506     
  Branches        0      932     +932     
==========================================
+ Hits         4570     7569    +2999     
- Misses        406      913     +507     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kintel]

Awesome - tested and works perfectly also with extra torture:

[kintel]

@elalish I've got another one. It's a bit time consuming to manually trim polygons to minimal examples. Do you have a good/preferred way of providing data-driven polygon tests?

[elalish]

If you compile with MANIFOLD_DEBUG and set ManifoldParams().processOverlaps = false, then it will error out when a bad triangulation is found and print out a test case to the command line that you can copy directly into our tests, like this.

[kintel]

After a while I figured out it's PolygonParams(), not ManifoldParams(). Will open a new ticket.

[elalish]

Oh, oops - you're right.