Create manifold STL meshes

[danieltwagner]

I made some headway on #18 - when interpolating points along the cube edges we bailed early if v1 or v2 was close to zero. However, if both v1 and v2 are close to zero then picking one or the other point means we might get unstable results depending on the order in which edges are evaluated.

This PR proposes to pick the midpoint between the two edges if there's no meaningful difference between the SDF values at the respective endpoints. Together with the script posted in #18 this helps with single-incident edges:

bottom-octtree.stl before:

Number of edges with 1 incident faces: 896
Number of edges with 2 incident faces: 963776
Number of edges with 3 incident faces: 5
Number of edges with 4 incident faces: 300
Number of edges with 5 incident faces: 3
Number of edges with 6 incident faces: 1565

bottom-octtree.stl after:

Number of edges with 2 incident faces: 964678
Number of edges with 4 incident faces: 82
Number of edges with 6 incident faces: 1564

[danieltwagner]

I added a second patch to omit faces with zero area (more specifically, triangles where at least one vertex is duplicated). With both patches applied we produce manifold STL meshes.

I would appreciate feedback on the approach and side-effects of the first patch (does it degrade any other functionality?) and perhaps you have thoughts around how to accomplish what the second patch does in a cleaner way. My understanding is this only happens if we pick the same extreme (=edge endpoint) for two edges of a face in mcInterpolate.

[deadsy]

Looks good - and apparently fixes the source of trouble for this particular model.