Differentiable cosmo: Cosmology now a full module with optional Phytorch

[mjyb16]

If I did not break anything regarding Caustic design principles, I think everything is ready to go. I tested creating a lens with the new Phytorch differentiable cosmology and it works, and the distances (especially gradients) are more accurate with Phytorch.

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.91% 🎉

Comparison is base (33cf6eb) 85.34% compared to head (4da6311) 86.25%.
Report is 13 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #72      +/-   ##
==========================================
+ Coverage   85.34%   86.25%   +0.91%     
==========================================
  Files          33       36       +3     
  Lines        1501     1608     +107     
==========================================
+ Hits         1281     1387     +106     
- Misses        220      221       +1     

Files Changed	Coverage Δ
caustic/cosmology/FlatLambdaCDM.py	100.00% <100.00%> (ø)
caustic/cosmology/LambdaCDM.py	100.00% <100.00%> (ø)
caustic/cosmology/__init__.py	100.00% <100.00%> (ø)
caustic/cosmology/base.py	84.21% <100.00%> (ø)

... and 3 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mjyb16]

All tests pass now that I have removed phytorch from the testing sequence. If people want to test phytorch, they can uncomment the test I wrote in test_cosmology.py, but this requires installing phytorch of course. Phytorch has been removed from the init.py inside the cosmology package, so Caustic will not try to import phytorch unless you explicitly import one of the cosmologies that uses it.

[mjyb16]

Current status: cannot build boost, which is needed to compile phytorch's cosmology.

[mjyb16]

Well, that's a wrap! Phytorch is now testable from within Caustic, with some caveats. The only real casualty is testing on windows....unfortunately, there is some problem with compiling C++ as usual on windows, and so I just decided to eliminate the windows-based Github actions. Who uses Windows anyway? Second thing is that I am relying on a couple of small modifications to phytorch's compilation process in order to help gcc find boost inside the github action, so I have forked phytorch. However, I will be syncing that fork regularly, and the changes are in a place where I don't think there will be any issues going forward. Thirdly, I have a file within the forked phytorch's home directory which I move into pytorch to overwrite a pytorch file which requires a small modification, at least until pytorch gets a pull request and fixes the issue. Fourthly, I had to reduce the accuracy of the critical density wrt to astropy to 1e-2, but I think this is due to floating point precision with very small numbers.

[ConnorStoneAstro]

This looks great! Just a few questions

[mjyb16]

All of Connor's requested changes have been implemented! Windows testing is back but without cosmology.