(fix): ndim accessible as np.ndim on PandasExtensionArray

[ilan-gold]

Reproducer for the underlying bug:

import pandas as pd
from xarray.core.extension_array import PandasExtensionArray
from xarray.core.indexing import LazilyIndexedArray

cat = PandasExtensionArray(pd.Categorical(["a", "b"] * 5))
lazy = LazilyIndexedArray(cat)
assert (lazy.get_duck_array().array == cat).all()

venv_13/lib/python3.13/site-packages/legacy_api_wrap/__init__.py:82: in fn_compatible
    return fn(*args_all, **kw)
src/anndata/_core/anndata.py:1433: in to_memory
    new[attr_name] = to_memory(attr, copy=copy)
/opt/homebrew/Cellar/[email protected]/3.13.0_1/Frameworks/Python.framework/Versions/3.13/lib/python3.13/functools.py:929: in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
src/anndata/_core/file_backing.py:168: in _
    return x.to_memory(copy=copy)
src/anndata/_core/xarray.py:203: in to_memory
    df = self.ds.to_dataframe()
venv_13/lib/python3.13/site-packages/xarray/core/dataset.py:7164: in to_dataframe
    return self._to_dataframe(ordered_dims=ordered_dims)
venv_13/lib/python3.13/site-packages/xarray/core/dataset.py:7086: in _to_dataframe
    if not is_extension_array_dtype(self.variables[k].data)
venv_13/lib/python3.13/site-packages/xarray/core/variable.py:416: in data
    duck_array = self._data.get_duck_array()
venv_13/lib/python3.13/site-packages/xarray/core/indexing.py:662: in get_duck_array
    return _wrap_numpy_scalars(array)
venv_13/lib/python3.13/site-packages/xarray/core/indexing.py:772: in _wrap_numpy_scalars
    if np.ndim(array) == 0 and (
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = PandasExtensionArray(array=['k', 'v', 'l', 'J', 't', ..., 'H', 'u', 'd', 'b', 'h']
Length: 1000
Categories (52, object): ['A', 'B', 'C', 'D', ..., 'w', 'x', 'y', 'z']), func = <function ndim at 0x101dfbe30>, types = (<class 'xarray.core.extension_array.PandasExtensionArray'>,)
args = (['k', 'v', 'l', 'J', 't', ..., 'H', 'u', 'd', 'b', 'h']
Length: 1000
Categories (52, object): ['A', 'B', 'C', 'D', ..., 'w', 'x', 'y', 'z'],), kwargs = {}

    def __array_function__(self, func, types, args, kwargs):
        def replace_duck_with_extension_array(args) -> list:
            args_as_list = list(args)
            for index, value in enumerate(args_as_list):
                if isinstance(value, PandasExtensionArray):
                    args_as_list[index] = value.array
                elif isinstance(
                    value, tuple
                ):  # should handle more than just tuple? iterable?
                    args_as_list[index] = tuple(
                        replace_duck_with_extension_array(value)
                    )
                elif isinstance(value, list):
                    args_as_list[index] = replace_duck_with_extension_array(value)
            return args_as_list
    
        args = tuple(replace_duck_with_extension_array(args))
        if func not in HANDLED_EXTENSION_ARRAY_FUNCTIONS:
>           raise KeyError("Function not registered for pandas extension arrays.")
E           KeyError: 'Function not registered for pandas extension arrays.'

Since #8821 we need ndim accessible via np.ndim and since https://github.com/pydata/xarray/pull/10317/files#diff-c803294f5216cbbdffa30f0b0c9f16a7e39855d4dd309c88d654bc317a78adc0L103-L104 we are not randomly falling back on whatever pandas offers. Hence this fix to explicitly add ndim to the numpy-registered functions.

• See https://github.com/scverse/anndata/actions/runs/15581661734/job/43878207681?pr=1997 our CI started failing on the last release.
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[ilan-gold]

Test failures looking unrelated

[dcherian]

How about we add ndim to duck_array_ops.py, try array.ndim first and then fall back to np.ndim to handle the scalar case?

[ilan-gold]

@dcherian Let me know if this is what you had in mind