Fix PandasArrayExtensionDtype comparison against strings

[cpruijsen]

Summary

Implement construct_from_string on PandasArrayExtensionDtype: parse array[<dtype>] into a PandasArrayExtensionDtype, and raise TypeError for anything else (how pandas spells "not my dtype" in __eq__). Provenance: pandas extension-dtype docs, and the reproduction on main in #8467 (pandas 3.0.5).

PandasArrayExtensionDtype.name is an instance property (array[float64]), so the inherited ExtensionDtype.construct_from_string asserts on a property object. Pandas compares dtypes to strings throughout its internals (is_string_dtype, astype_is_view), so any Array2D/3D/4D/5D column that reaches pandas raises AssertionError on dtype == "string", is_string_dtype, and df.astype(object).

This PR does not change _metadata from "value_type" to ("value_type",). That is #8375 / #8464, a distinct contract bug.

Landing construct_from_string and the _metadata tuple in one PR would also stop dtype == "array[float64]" (the dtype's own name) from raising. With only this change, dtype == "string" is False and df.astype(object) works; own-name comparison still hits the _metadata AttributeError. Can fold the tuple in if you would rather land them together.

Fixes #8467

Test plan

• tests/features/test_array_xd.py construct-from-string round-trip (array[int32] / array[bool] / array[float64])
• construct_from_string raises TypeError for string, int64, array[], array[not_a_dtype], not_an_array[int32]
• test_table_to_pandas_dtype_compares_to_string: dtype == "string" is False, is_string_dtype is False, df.astype(object) works
• The 11 new cases fail on main (AssertionError: (PandasArrayExtensionDtype, <class 'property'>)) and pass here; full tests/features/test_array_xd.py is 114 passed