levelset(): let edge be one spacing per axis

[revarbat]

Refs BelfrySCAD/BelfrySCAD#729

edge=[ex,ey,ez] (or [ex,ey] for a 2D bounds) samples each axis at its own spacing — the equivalent of BOSL2 isosurface()'s vector voxel_size, which the reporter needs to use levelset() as a drop-in for BOSL2's marching cubes.

How: Manifold::LevelSet takes a single edge length, so the 3D path meshes the field squashed to unit spacing on every axis and Scales the mesh back. This is exact, not an approximation: with tolerance −1 each vertex is a linear interpolation along a lattice edge, and that commutes with an axis scale. AVectorEdgeIsExactlyAStretchedLattice pins it: a sphere stretched 4× in z, sampled 4× coarser in z, gives the same triangle count and exactly 4× the volume (1e-6). The 2D path already had separate x/y spacings, so it just reads them from the vector.

A plain number takes the old code path unchanged, so existing scripts mesh identically.

Also fixed on the way: a non-positive edge= given with a grid field warned, then left edge at 0 and passed that to LevelSet; it now falls back to the grid spacing.

Feature level: supported_feature("levelset") is now 2, so a library can probe for a vector edge. FeatureDetection.EveryDocumentedFeatureReportsALevel checked for the digit 1 in the echo; it now checks >= 1.

closed=false (also asked for in the issue) is not included: it means an open surface, which can't be a solid.

Tests: 6 new LevelSetFn cases; full suite 1485 passed, 0 failed.

🤖 Generated with Claude Code