feat: add bbknn

[Intron7]

This PR add a scanpy native bbknn implementation because of the removal of external

fixes #4302

[codecov]

Codecov Report

❌ Patch coverage is 94.11765% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.14%. Comparing base (6b5405d) to head (5fa4dd2).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/scanpy/neighbors/_common.py	79.48%	8 Missing ⚠️
src/scanpy/neighbors/_bbknn.py	98.85%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4306      +/-   ##
==========================================
+ Coverage   80.01%   80.14%   +0.13%     
==========================================
  Files         132      133       +1     
  Lines       13397    13511     +114     
==========================================
+ Hits        10719    10828     +109     
- Misses       2678     2683       +5     

Flag	Coverage Δ
hatch-test.low-vers	78.38% <94.11%> (+0.14%)	⬆️
hatch-test.pre	80.01% <94.11%> (+0.13%)	⬆️

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	75.79% <ø> (ø)
src/scanpy/external/pp/_bbknn.py	45.45% <ø> (ø)
src/scanpy/neighbors/__init__.py	81.49% <100.00%> (-0.84%)	⬇️
src/scanpy/neighbors/_types.py	100.00% <100.00%> (ø)
src/scanpy/preprocessing/__init__.py	100.00% <100.00%> (ø)
src/scanpy/neighbors/_bbknn.py	98.85% <98.85%> (ø)
src/scanpy/neighbors/_common.py	70.83% <79.48%> (+5.92%)	⬆️

... and 1 file with indirect coverage changes

[flying-sheep]

Looks great!

I deduplicated the transformer creation method and adapted the heuristic a bit, and added a benchmark suite to check it. Take a look if that makes sense or if I missed something!

The only piece I think is missing is that we should get rid of legacy cruft, but I can take care of that:

• no reason for _accepts_legacy_random_state
• things like batch_key and so on should become something accepting AdRefs or strs in the resolve format ("obs.obs_colname" instead of "obs_colname", and then _key of course makes no sense).