build!: drop Python 3.11 and upgrade the spatialdata cluster

[Tomatokeftes]

Closes #117. Shrinks #114.

Stacked on #119. Base it there or merge #119 first; the diff shown here
includes #119's two commits.

This is a breaking change and the commit is marked BREAKING CHANGE:, so
semantic-release will cut 2.0.0. If you would rather ship it as a minor,
reword the footer before merging.

Why

spatialdata 0.8.0 and anndata 0.13 both require Python >= 3.12. That single
constraint was blocking the entire dependency cluster, and with it the removal
of the pandas 3 string-dtype coercion. Dropping 3.11 unblocks all of it at
once.

Pins

package	before	after	resolved
python	>=3.11, <3.14	>=3.12, <3.14	3.13
dask	>=2025.12.0, <2026.2	>=2026.3.0, <2026.8	2026.7.1
spatialdata	>=0.7.3, <0.8	>=0.8.0, <0.9	0.8.0
ome-zarr	>=0.14.0, <0.16	>=0.16.0, <0.19	0.18.0
anndata	>=0.11.0, <0.13	>=0.13.2, <0.14	0.13.2
zarr	>=3.0.0, <3.2	unchanged	3.1.3

zarr did not need to move, which was not obvious going in.

The anndata floor is 0.13.2, not 0.13.0, deliberately. anndata 0.13.0
imports typing_extensions.sentinel (needs typing-extensions >= 4.16) but
declares no floor for it. A resolver therefore prefers 0.13.0 -- looser
metadata is easier to satisfy -- pairs it with typing-extensions 4.14.1, and
produces an environment that installs cleanly and then dies at import:

File "anndata/acc/__init__.py", line 13, in <module>
    from typing_extensions import sentinel
ImportError: cannot import name 'sentinel' from 'typing_extensions'

A floor of >=0.13 would have shipped that. 0.13.2 declares it correctly.
This is exactly the failure mode the clean-venv CI job exists to catch.

What the coercion removal rests on

_coerce_table_strings_to_object existed because anndata could not serialize
pandas' str dtype. anndata 0.13.0 fixed that (anndata #2221, via #2133).
Verified directly before removing it -- writing an obs frame with a string
index, a string column and a string-backed categorical:

pandas	pyarrow	anndata	write_zarr
2.3.2 + infer_string	yes	0.12.2	IORegistryError
3.0.5	yes	0.12.2	IORegistryError
2.3.2 + infer_string	yes	0.13.2	PASS
3.0.5	yes	0.13.2	PASS

Tests

pass	main	this branch
unit	643 passed, 1 xfailed	641 passed
integration	4 failed, 14 passed	4 failed, 14 passed

641 = 643 - 3 deleted coercion tests + 1 converted xfail. The four integration
failures are identical to main and unrelated to this change: three tests pass a
'linear' resampling method the code now rejects, one is a COO region
mismatch. They were failing before this PR and still are.

Test changes:

• Deleted the three tests asserting the coercion's own behaviour.
• Kept the write-path tests under future.infer_string=True for all three
  paths. They now cover anndata's handling rather than Thyra's workaround, and
  nothing else in the suite runs with string inference on.
• Converted the xfail removal trigger into a plain assertion, so a regression
  in anndata's string support is still caught at the smallest scope.

mkdocs --strict clean. black / isort / flake8 / mypy / bandit / pydocstyle
clean via pre-commit.

Also updated

CI matrices in tests.yml, release.yml and complexity-monitoring.yml;
classifiers; [tool.mypy] python_version; and the docs and README badges that
advertised 3.11 -- several of which said "3.11 or 3.12" and were stale already,
since 3.13 has been supported for a while.

The dependency-cluster comments are re-pointed too: the pandas comment cited a
coercion that no longer exists, the ome-zarr comment explained a ceiling that
has moved, and the cluster preamble used a worked example that is now the
combination we ship.

Known, not introduced here

A repo-wide poetry run mypy thyra reports 30 errors across 9 files. None are
in code this PR touches, and none are gated: mypy is not in CI at all, and the
pre-commit hook runs with --ignore-missing-imports --no-strict-optional in an
isolated env that cannot see spatialdata or anndata. Some of the union-attr
signals likely widened because anndata 0.13 types X more richly. Worth a
follow-up, out of scope here.

Follow-up for #114

This moves the spatialdata ceiling to 0.8.0. When scverse/spatialdata#1055
eventually lands, #114 is reduced to a floor bump rather than the whole cluster
migration it describes today.

[github-actions]

Complexity Monitoring Report

Threshold: 15
Total Violations: 0

Excellent! No complexity violations found.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!