Add Crispat Guideassignment

[Intron7]

Here we will add Guideassignment to pertpy GPU based on what Crispat does

[codecov-commenter]

Codecov Report

❌ Patch coverage is 74.87437% with 50 lines in your changes missing coverage. Please review.
✅ Project coverage is 88.48%. Comparing base (32c424a) to head (d97fa49).

Files with missing lines	Patch %	Lines
.../rapids_singlecell/pertpy_gpu/_guide_assignment.py	74.74%	50 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #637      +/-   ##
==========================================
- Coverage   88.83%   88.48%   -0.36%     
==========================================
  Files         100      101       +1     
  Lines        7615     7814     +199     
==========================================
+ Hits         6765     6914     +149     
- Misses        850      900      +50     

Files with missing lines	Coverage Δ
src/rapids_singlecell/_cuda/__init__.py	80.00% <ø> (ø)
src/rapids_singlecell/pertpy_gpu/__init__.py	100.00% <100.00%> (ø)
.../rapids_singlecell/pertpy_gpu/_guide_assignment.py	74.74% <74.74%> (ø)

[Zethson]

Thank you very much! Awesome stuff.

1. Maybe it's just because it's the first time I'm seeing this but I'm confused that there's so many hasattr checks for methods to get the (dense & sparse) device arrays to the host in tests. There should be a trivial utility function for this like https://rapids-singlecell.readthedocs.io/en/latest/api/generated/rapids_singlecell.get.X_to_CPU.html
2. We should also show this off in a tutorial. Could you please also make a PR to the tutorials repository?
3. I always like it when there's small runnable examples in the PR description. We're linking to PRs in the changelog so when people look at new features, they should get an improved PR description. Also something your LLM can do if you have gh set up.