add velocity infrastructure.

[MFraters]

This pull request implements #528. It is still very rudementary and a work in progress.

I am thinking of renaming the current uniform models to raw, since they just work with raw velocity input. I think the uniform model should be more useful, given that it is usually the simple default model. I think uniform in this case should not make the same value everywhere (which is would raw (constant?) would be, but for example uniform in continent would provide a direction, dip and magnitude (which is more like how the rest of the models work, 2d top down + dip) and for a slab or fault it would be relative to the slab geometry. In spherical uniform should also be relative to the geometry.

Example of current functionality:

[github-actions]

Benchmark	Main	Feature	Difference (99.9% CI)
Slab interpolation simple none	0.992 ± 0.005 (s=430)	1.194 ± 0.011 (s=399)	+20.1% .. +20.5%
Slab interpolation curved simple none	0.993 ± 0.004 (s=399)	1.195 ± 0.013 (s=424)	+20.1% .. +20.5%
Spherical slab interpolation simple none	1.161 ± 0.008 (s=369)	1.209 ± 0.011 (s=392)	+3.9% .. +4.3%
Slab interpolation simple curved CMS	1.038 ± 0.005 (s=403)	1.235 ± 0.012 (s=392)	+18.7% .. +19.1%
Spherical slab interpolation simple CMS	1.566 ± 0.010 (s=267)	1.624 ± 0.015 (s=299)	+3.5% .. +3.9%
Spherical fault interpolation simple none	1.166 ± 0.009 (s=384)	1.215 ± 0.011 (s=375)	+4.0% .. +4.4%
Cartesian min max surface	2.347 ± 0.045 (s=165)	2.707 ± 0.023 (s=192)	+14.8% .. +15.9%
Spherical min max surface	7.031 ± 0.026 (s=81)	7.637 ± 0.113 (s=46)	+7.8% .. +9.5%

[MFraters]

This is almost ready to merge I think. It also contains some other items, including an option to replace the reference results from any test that fails, and limiting the resolution of the doc_* tests to improve tester performance.