Repair meshes in manifold & fast-csg routes

[ochafik]

This PR introduces some crude mesh repairs in manifold & fast-csg modes.

Since orient_to_bound_a_volume seems happy to throw assert and precondition exceptions at us when the mesh has self-intersections or isn't closed, I'm catching these, remove any self intersections & plug any holes (then try orienting again w/ a more lenient method).

This fixes 3D/features/linear_extrude-scale-zero-tests.scad along with a few models I had that have invalid geometry but that I prefer seeing repaired than exploding in flight.

TODO before undrafting:

• Are unclosed meshes ever valid? This PR is probably enforcing their repair / closing, with extra costs (before, orientation routine was fenced by CGAL::is_closed)
• Check performance regression (switched to using Epick kernel in Manifold factory to avoid inexact checks in orientation, but this could be slower)
• Add more tests?

[pca006132]

You can use the hull function in manifold. We don't do much optimization for now but we may do them later.

[pca006132]

Oh sorry, I probably checked the wrong thing.

[ochafik]

Hah, good to know it's there, not sure about here (would have to inline this logic in a manifold-specific way) but I'll probably give it a spin where we perform the hull primitive itself.

[pca006132]

Looking at CGAL, it seems that they do have algorithm for repairing non-manifold triangular soup called duplicate_non_manifold_vertices. They are called combinatorial repair https://doc.cgal.org/latest/Polygon_mesh_processing/index.html#PMPCombinatorialRepair.

I think the existing mesh processing in this PR are for dealing with self-intersections, but not sure if this can deal with topological issues (manifoldness).

[pca006132]

Ah, bad news. I tried duplicate_non_manifold_vertices and repair_polygon_soup with no luck. It doesn't work on the example in #4838.

[pca006132]

Good news: We need to first repair_polygon_soup, and then orient_polygon_soup. Here is a demo:

782eb0c

This works for the simple cases where two cubes are touching at a corner/edge. I did not try more complicated cases. Should be nice to include in this PR.

That code is basically copied from https://github.com/CGAL/cgal/blob/master/Polygon_mesh_processing/include/CGAL/Polygon_mesh_processing/IO/polygon_mesh_io.h