feat: custom in-memory indexes

[ilan-gold]

From the proteomics hackathon, we definitely need to define some parameters and work out kinks:

1. Writeability: feat: AnnData.unwriteable based on AnnData._reduce + iter_outer + refactorings of other relevant functions #2372 + Lenient write #2373 + Handle new zarr dtypes (both custom and datetime) #2238. Conretely, feat: AnnData.unwriteable based on AnnData._reduce + iter_outer + refactorings of other relevant functions #2372 needs to go in first to prevent people from even trying to write these i.e., when there is a MultiIndex
2. What exactly are we changing? In other words, what does this setting do exactly? I think the two tests are clear about this, but there is some subtle stuff like you used to be able to do adata.obs_names = non_string_index and this evidently worked? This will be essential for documentation
3. Do we need any guardrails? No arrow integer types? What sort of filtering can/should we do here?
4. In the case of no index provided on declaration, should we make new decisions? https://github.com/scverse/pandas-uuid maybe?

• Towards i/o of different index types #2241 and Feature request - var_names/obs_names as fixed-sized types (integer or bytes) #777
• Tests added

• Release note not necessary because:

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.60%. Comparing base (9f2fa25) to head (ae1c935).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2379      +/-   ##
==========================================
+ Coverage   85.57%   85.60%   +0.02%     
==========================================
  Files          49       49              
  Lines        7669     7671       +2     
==========================================
+ Hits         6563     6567       +4     
+ Misses       1106     1104       -2     

Files with missing lines	Coverage Δ
src/anndata/_core/aligned_df.py	96.72% <100.00%> (+0.05%)	⬆️
src/anndata/_core/anndata.py	86.81% <100.00%> (ø)
src/anndata/_core/index.py	95.23% <100.00%> (+0.69%)	⬆️
src/anndata/_io/read.py	52.79% <100.00%> (+0.24%)	⬆️
src/anndata/_settings.py	100.00% <100.00%> (ø)
src/anndata/tests/helpers.py	83.88% <100.00%> (ø)

[ilan-gold]

Just looking for preliminary thoughts here @flying-sheep

[flying-sheep]

direction looks great!

[ilan-gold]

What exactly are we changing? In other words, what does this setting do exactly? I think the two tests are clear about this, but there is some subtle stuff like you used to be able to do adata.obs_names = non_string_index and this evidently worked? This will be essential for documentation

This is still an outstanding question to me but on second thought, nothing seems to have actually changed. This was always allowed before, but maybe this PR should disallow it if the restriction setting is on? There's a warning about this but that's it. So maybe a test would be good

UPDATE: Ok I thought of that and https://github.com/scverse/anndata/pull/2379/changes#diff-93807bfddb872b9cf4ea7586b7ee8feb73bb9cab28969a82b4b6974001ff3462R749-R753 is what we do i.e., leave the old mechanism in place, which is probably the least disruptive thing to do

[flying-sheep]

Awesome! Very nice to see so few changes necessary, we really improved this codebase!

/edit: GitHub however did not, I had to delete and recreate the comment below, since GitHub inexplicably moved it to a different line, breaking the diff.