Use RegularGridInterpolator for proper extrapolation of small PSF images

[oczoske]

This fixes #797 .

• The interpolator has been changed from RectBilinearSpline to RegularGridInterpolator. While the interface is more cumbersome it has better support for extrapolation - here we need fill_value=0. We still use linear interpolation but might consider using pchip; tests are needed.
• The spatial size of the scaled PSF cube is at most twice the slit size (ensures full coverage of the slit by the psf), otherwise it takes the size of the input PSF image. This can be further reduced by setting the parameter cmd["!SIM.computing.psf_maxsize"]; this can be done to improve computation speed at the expense of accuracy.

[codecov]

Codecov Report

❌ Patch coverage is 14.28571% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.88%. Comparing base (e161381) to head (f61afe8).
⚠️ Report is 146 commits behind head on main.

Files with missing lines	Patch %	Lines
scopesim/effects/psfs/discrete.py	14.28%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #798      +/-   ##
==========================================
+ Coverage   74.86%   74.88%   +0.01%     
==========================================
  Files          69       69              
  Lines        8627     8628       +1     
==========================================
+ Hits         6459     6461       +2     
+ Misses       2168     2167       -1     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[teutoburg]

Do we know how this compares in terms of performance? The previous RectBilinearSpline always came up in profiling, so it would be interesting if RegularGridInterpolator is any better or worse, or in the same ballpark. I'm fine with a slightly worse performance if the tradeoff is (much) better extrapolation and thus more accurate results, but if it's something like an order of magnitude worse, we'd have to consider alternatives. If it turns out it actually improves performance, all the better!