Allow writing StringDType variables to netCDF

[kkollsga]

• Closes Datasets concatenated along string dimension cannot write to netCDF #11199
• Tests added
• User visible changes documented in whats-new.rst
• New functions/methods listed in api.rst (N/A — no new public API)

Summary

• Recognizes numpy.dtypes.StringDType (kind "T") as a unicode string type in is_unicode_dtype, so the encoding pipeline and backend dtype selection handle it correctly.
• Converts StringDType arrays to object arrays in netCDF4 and h5netcdf backend prepare_variable methods, since neither C library supports StringDType natively.
• Null values from StringDType(na_object=None) are replaced with empty strings on write, matching existing behavior for object-dtype string arrays with missing values.
• The scipy backend already works because EncodedStringCoder(allows_unicode=False) encodes strings to bytes via encode_string_array, which handles StringDType.

Test plan

• test_is_unicode_dtype_stringdtype — unit test for is_unicode_dtype with StringDType
• test_roundtrip_stringdtype_data — roundtrip test in DatasetIOBase, runs across all backends (netCDF4, h5netcdf, scipy, zarr)
• Manual verification of null handling with StringDType(na_object=None)
• Pre-commit (ruff, formatting) passes
• mypy passes (no new errors)

🤖 Generated with Claude Code

[kkollsga]

The mypy and test failures look related to numpy 2.4.2's stricter type stubs (#11183, #11204).