Change strategy for computing lensing lightcones and adds tests for Density planes and Born convergence

[EiffL]

This PR changes a little bit how we are doing things in the lensing branch, but it should validate almost all we care about.

• The main modification is that I realized duplicating the nbody code in tfpm and raytracing was not a great idea, so I just added an option to export the intermediate steps of the nbody and we can use that to access the simulation volume at the scale factors we care about.
• Given a simulation volume/snapshot, I implemented an equivalent of the cutPlaneGaussianGrid of lentools, and validated it in tests/test_raytracing.py:test_density_plane This makes sure that we are in 100% agreement at the level of the density planes
• Then I implemented a convergenceBorn function that works similarly to the lenstools one, and this one is validated in tests/test_raytracing.py:test_convergence_Born which compares the convergence maps for a given set of density planes.

You can see an example of this proposed new API here: https://github.com/DifferentiableUniverseInitiative/flowpm/blob/u/EiffL/lensing/notebooks/lensing.ipynb

And I have also added a dev notebook to see the outputs of these test functions here: https://github.com/DifferentiableUniverseInitiative/flowpm/blob/u/EiffL/lensing/notebooks/dev/test_raytracing.ipynb

Highlights of comparisons

Density planes

This is comparison of the density planes from the same snapshot, note that the high freq are slightly different because of the different painting schemes, but the residuals are clean. After smoothing, we validate these maps pixel-wise at a maximum relative error of 2%, their smoothed power spectra are also good:

Convergence maps

This is the output of running an LPT simulation, cutting it in density planes using FlowPM, and then doing the ray tracing from these planes in both lenstools and FlowPM. Very good match, validated at an absolute tolerance of 5e-4

[EiffL]

Let me know what you think @dlanzieri but I think this may be a full validation against lenstools. Now the only thing that remains to be seen is whether we can recover a reasonable power spectrum....

[EiffL]

Thank you @dlanzieri :-D since you approve these modifications, I'll go ahead and merge to the lensing branch.