Clean up intraband test
Convert the infraband test from pure garbage to something reasonable.
- Move gpw generation to fixture so we save time in most runs (factor 4 speedup once gpws are cached)
- Use a helper class to compare dielectric function calculations
I also note that the test doesn't break even if I moderately 'sabotage' the implementation to produce wrong numbers. So there should probably be a proper convergence test to verify that tetrahedron and point integration actually converge to the same result in an actual unit test.