cufinufft v2.2.0 migration

[garrettwrong]

This updates our pyproject.toml towards CUDA 12 with cufinufft v2.2.0.

For cufinufft v2.2.0, the changes described in the python migration guide were applied.

I was expecting this release to solve the float/double bug originally reported in the legacy cufinufft repo, but it seems like that bug (or one with the same symptoms) is still there in the pypi package. I will look into this just a little and potentially create a bug report in the main finufft repo.

I'm not sure why this seg faulted in the CI. I ran into this locally, but it went away after some (system) package changes. Perhaps I need to restart the runner service. Will look into that.

[codecov]

Codecov Report

Attention: 7 lines in your changes are missing coverage. Please review.

Comparison is base (39d269a) 88.82% compared to head (1a4fab7) 88.82%.

Files	Patch %	Lines
src/aspire/nufft/cufinufft.py	0.00%	7 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #1087   +/-   ##
========================================
  Coverage    88.82%   88.82%           
========================================
  Files          126      126           
  Lines        12028    12028           
========================================
  Hits         10684    10684           
  Misses        1344     1344           

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

[garrettwrong]

The seg fault occurred because with the new packages/env pip is not handling _ in package extension names the same way. This meant that the gpu package extension was not found, and the current cufinufft was not installed in the CI env.

I've converted the package extensions to use - (what is prefered now apparently) in the pyproject.toml and documentation.

[j-c-c]

Looks good!

[garrettwrong]

I'm going to push this through to clear up the CI.