Implement __arrow_array__ on PandasArrayExtensionArray so ArrayXD columns can go back from pandas to Arrow

[JoeyTan21]

Fixes #8699

Dataset.to_pandas() returns ArrayXD columns as PandasArrayExtensionArray, but that extension array never implemented pyarrow's __arrow_array__ protocol, so pa.Table.from_pandas raised ArrowTypeError: Did not pass numpy.dtype object. This broke Dataset.from_pandas(ds.to_pandas()) and ds.with_format("pandas").map(fn, batched=True) for any dataset with an Array2D..Array5D column.

This PR adds PandasArrayExtensionArray.__arrow_array__:

• builds the nested list storage with the existing to_pyarrow_listarray helper (same code path as ArrowWriter), so fixed-shape and dynamic-first-dimension arrays and null rows are handled the same way as when writing;
• wraps it in the matching ArrayNDExtensionType, taken from the requested type when pyarrow passes one (e.g. schema=), otherwise inferred from the numpy data (shape/dtype), so the column keeps its ArrayXD feature after the round trip;
• if a non-extension Arrow type is requested, returns the storage cast to it.

Tests: test_array_xd_from_pandas (fixed + dynamic shape; covers pa.array(series.array), Dataset.from_pandas with/without features, and a pandas-formatted batched map) and test_array_xd_from_pandas_with_none (null row preserved as an Arrow null). tests/features/test_array_xd.py, test_features.py, test_formatting.py and the pandas/map subset of test_arrow_dataset.py pass; ruff check / ruff format --check clean on the changed files.

Disclosure: found, reproduced and the draft fix tested locally with the help of an AI coding agent (Claude Code); reviewed before filing.

🤖 Generated with Claude Code

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.