(feat): new zarr dtypes

[ilan-gold]

• Accounting for refactor v3 data types zarr-developers/zarr-python#2874
• Tests added
• Release note added (or unnecessary)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.59%. Comparing base (e20456a) to head (bbf7507).
⚠️ Report is 49 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1995      +/-   ##
==========================================
- Coverage   87.59%   85.59%   -2.00%     
==========================================
  Files          46       46              
  Lines        7062     7054       -8     
==========================================
- Hits         6186     6038     -148     
- Misses        876     1016     +140     

Files with missing lines	Coverage Δ
src/anndata/_core/xarray.py	98.42% <100.00%> (ø)
src/anndata/_io/specs/methods.py	90.69% <100.00%> (-0.02%)	⬇️
src/anndata/_io/zarr.py	83.90% <ø> (-0.88%)	⬇️
src/anndata/compat/__init__.py	80.73% <100.00%> (ø)
src/anndata/experimental/backed/_lazy_arrays.py	94.23% <ø> (-0.06%)	⬇️
src/anndata/tests/helpers.py	83.20% <100.00%> (-10.56%)	⬇️

... and 5 files with indirect coverage changes

[flying-sheep]

Very nice that this now gets more straightforward!

Seems like zarr.dtype.VariableLengthUTF8 wraps numpy’s new string dtype if available. Would be cool if you explained such things in the PR description.

[ilan-gold]

Seems like zarr.dtype.VariableLengthUTF8 wraps numpy’s new string dtype if available.

Hmm, I hadn't thought of it that way tbh, it was just "the new zarr API for dtypes" and this happens to be the only place we use it. I will give more context next time, apologies!

[flying-sheep]

Makes sense! I just wondered what that dtype is and found that it’s basically just our previous match statement, but done by zarr now.

[ilan-gold]

More info here! https://zarr.readthedocs.io/en/stable/api/zarr/dtype/index.html This PR is basically what I held up the anndata release over - the new dtype stuff was big enough that I wanted a clean break. But it is a very rich feature set! I haven't had time to dig into the new possibilities yet, but looks promising