(chore): clearer handling of consolidated metadata

[ilan-gold]

I noticed in https://github.com/scverse/anndata/pull/1995/files#diff-0edf267bfe4d54327694a6b3a05f6035bd138af62140d418bc4e01b11634028bR143-R150 that handling consolidated metadata can be a bit tricky, so I make things clearer in code and documentation what should be expected. I subscribe to pydata/xarray#10122 (comment) that without the feature, cloud data is unusably slow. I would ideally like to allow folks to opt out of consolidated using zarr-developers/zarr-python#3119 instead of an argument to the write_zarr function but the timeline on the release of this feature is unclear at the moment.

• Closes #
• Tests added
• Release note added (or unnecessary)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.47%. Comparing base (7821a27) to head (2466d8f).
⚠️ Report is 44 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2015      +/-   ##
==========================================
- Coverage   87.45%   85.47%   -1.98%     
==========================================
  Files          46       46              
  Lines        7057     7066       +9     
==========================================
- Hits         6172     6040     -132     
- Misses        885     1026     +141     

Files with missing lines	Coverage Δ
src/anndata/_io/specs/registry.py	93.96% <100.00%> (-0.25%)	⬇️

... and 6 files with indirect coverage changes