3d material-grid test

[stevengj]

Probably the smallest reasonable test would be something like what @oskooi did in the subpixel-smoothing papers: use a small unit cell periodic in all 3 dimensions, with e.g. a dielectric sphere, and compare the eigenfrequency (as returned by harminv) to the same structure with a material grid (with or without subpixel smoothing).

Any 3d test will still be relatively expensive, so we might not want it to run in CI, but have a make bigcheck option that includes it which we run occasionally (e.g. before releases).