feat: Support pandas 3.0 upcoming changes

[flying-sheep]

TODO:

• save storage in data as well?
• make this numpy 1.x compatible so we can release it in 1.12.x?
• move following plan into issue(s) and add the second point to the 0.13 milestone

1. in this PR, make sure we support pd.options.future.infer_string=True/pandas>=3 by adding tests for everything that might break with future.infer_string=True together with (implicitly or explicitly) allow_write_nullable_strings set, and augmenting the error message that’s shown if the latter is False
2. in anndata 0.13’s release note, we add a fat warning that this setting will flip in 0.14, and if pandas’ version >=3, change allow_write_nullable_strings=False to mean “try writing them as non-nullable and fail only if we can’t”
3. in anndata 0.14 we flip allow_write_nullable_strings’ default to True

This activates the following settings as described in https://pandas.pydata.org/docs/whatsnew/v2.3.0.html

future.infer_string = True

This one causes many test failures we need to fix.

Mostly because ArrowStringArrayNumpySemantics isn’t a pd.arrays.StringArray

>>> from pandas.core.arrays.string_arrow import ArrowStringArrayNumpySemantics
>>> ArrowStringArrayNumpySemantics.mro()
[pandas.core.arrays.string_arrow.ArrowStringArrayNumpySemantics,
 pandas.core.arrays.string_arrow.ArrowStringArray,
 pandas.core.strings.object_array.ObjectStringArrayMixin,
 pandas.core.arrays.arrow.array.ArrowExtensionArray,
 pandas.core.arraylike.OpsMixin,
 pandas.core.arrays.base.ExtensionArraySupportsAnyAll,
 pandas.core.arrays.string_.BaseStringArray,
 pandas.core.arrays.base.ExtensionArray,
 pandas.core.arrays._arrow_string_mixins.ArrowStringArrayMixin,
 pandas.core.strings.base.BaseStringArrayMethods,
 abc.ABC,
 object]

mode.copy_on_write = True

Weirdly, this doesn’t cause any test failures! According to the docs it implies mode.chained_assignment = "raise", but the warning we suppressed in #545 doesn’t get elevated to an exception. And tests like test_base.py::test_slicing_* still pass. Maybe everything is fine?

We could also do pd.options.mode.copy_on_write = "warn" as the docs say …

[codecov]

Codecov Report

❌ Patch coverage is 52.50000% with 19 lines in your changes missing coverage. Please review.
✅ Project coverage is 79.72%. Comparing base (9da064f) to head (52212bb).

Files with missing lines	Patch %	Lines
src/anndata/experimental/backed/_lazy_arrays.py	27.27%	8 Missing ⚠️
src/anndata/_io/specs/methods.py	76.19%	5 Missing ⚠️
src/anndata/_core/merge.py	0.00%	4 Missing ⚠️
src/anndata/_io/specs/lazy_methods.py	0.00%	1 Missing ⚠️
src/anndata/compat/__init__.py	66.66%	1 Missing ⚠️

❌ Your project check has failed because the head coverage (79.72%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage.

❗ There is a different number of reports uploaded between BASE (9da064f) and HEAD (52212bb). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (9da064f)	HEAD (52212bb)
	3	2

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2133      +/-   ##
==========================================
- Coverage   85.42%   79.72%   -5.70%     
==========================================
  Files          46       46              
  Lines        7081     7098      +17     
==========================================
- Hits         6049     5659     -390     
- Misses       1032     1439     +407     

Files with missing lines	Coverage Δ
src/anndata/_core/anndata.py	83.18% <ø> (-0.13%)	⬇️
src/anndata/_core/views.py	85.40% <ø> (-0.43%)	⬇️
src/anndata/_settings.py	94.33% <ø> (ø)
src/anndata/utils.py	86.22% <ø> (ø)
src/anndata/_io/specs/lazy_methods.py	73.71% <0.00%> (-22.44%)	⬇️
src/anndata/compat/__init__.py	77.98% <66.66%> (-3.04%)	⬇️
src/anndata/_core/merge.py	74.05% <0.00%> (-11.06%)	⬇️
src/anndata/_io/specs/methods.py	83.18% <76.19%> (-6.69%)	⬇️
src/anndata/experimental/backed/_lazy_arrays.py	38.01% <27.27%> (-53.59%)	⬇️

... and 11 files with indirect coverage changes