BUG: Interchange Column.null_count is a NumPy scalar, not a builtin int

[honno]

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of pandas.
• I have confirmed this bug exists on the main branch of pandas.

In PandasColumn used for the dataframe interchange protocol, null_count returns the result of Series.sum(), which defaults to returning a 0d ndarray of the default integer (specifically a NumPy scalar). This is opposed to int, which is specified for null_count in the spec.

>>> df = pd.DataFrame({"foo": [42]})
>>> interchange_df = df.__dataframe__()
>>> interchange_col = interchange_df.get_column_by_name("foo")
>>> interchange_col.null_count
0
>>> type(interchange_col.null_count)
numpy.int64  # should be Python's int

Relevant code

pandas/pandas/core/exchange/column.py

Lines 180 to 184 in a7c5773

	def null_count(self) -> int:
	"""
	Number of null elements. Should always be known.
	"""
	return self._col.isna().sum()

This doesn't seem to be affecting interchange thus far, so might be a low-prio or wont-fix situation. I'd just add these 0d array situations has often tripped up logic assuming a Python int on the Array API side of things, and indeed there's currently a similiar issue for vaex's interchange implementation in vaexio/vaex#2093.

Using local build of main upstream.

[jbrockmendel]

No idea if it's relevant, but a np.int64 object is a scalar, which is distinct from a 0d numpy ndarray