Backport PR #4082 on branch 1.12.x (fix: limit Numba threads in Wilcoxon path of rank_genes_groups)

[flying-sheep]

Backport PR #4082

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.49%. Comparing base (cc62733) to head (e662e5f).
⚠️ Report is 11 commits behind head on 1.12.x.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##           1.12.x    #4088      +/-   ##
==========================================
+ Coverage   78.45%   78.49%   +0.04%     
==========================================
  Files         118      120       +2     
  Lines       12608    12632      +24     
==========================================
+ Hits         9892     9916      +24     
  Misses       2716     2716              

Flag	Coverage Δ
hatch-test.low-vers	77.79% <100.00%> (+0.04%)	⬆️
hatch-test.pre	78.38% <100.00%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/_utils/__init__.py	78.63% <100.00%> (+0.04%)	⬆️
src/scanpy/_utils/_numba.py	100.00% <100.00%> (ø)
src/scanpy/tools/_rank_genes_groups.py	92.77% <100.00%> (+0.02%)	⬆️

... and 8 files with indirect coverage changes

[zboldyga]

hey @flying-sheep I was just doing some benchmarking of another change I'm suggesting to the library for wilcoxon DE and I noticed the default behavior of the library has changed (as compared with published v 1.12.0 -- it defaults to using a single thread for wilcoxon DE now when n_jobs is unspecified). I think it's this PR

I believe the prior default was to use all available threads, which seems to make more sense in the default case?