ENH: implement Kleene versions of GroupBy any/all kernels for nullable dtypes

[arw2019]

xref #37493 (comment)

[phofl]

Wanted to look into this and tried to understand the implementation for Keene logic in case of any/all for a Series. This means, that

s = pd.Series([False, pd.NA, False], dtype="boolean")
print(s.any(skipna=False))

works and returns pd.NA. But

df= pd.DataFrame({"a": [False, pd.NA, False]}, dtype="boolean")
print(df.any(skipna=False))

returns an empty Series. I could not find the reason, why this was not implemted for DataFrames too, when any columns there is from dtype boolean. Should this get implemented for SeriesGroupBy and DataFrameGroupBy?

[arw2019]

I haven't gone through the whole path put I think we may want to do this in the cython method

[phofl]

For GroupBy that would probably be the best. But we have to fix #37505 before doing this, because this raises for

s = pd.Series([False, pd.NA, False], dtype="boolean", index=[0,0, 1])
grouped = s.groupby(level=0)
print(grouped.any(skipna=False))

too before even reaching the cython part. But I meant the regular DataFrame.any() without groupby. This does not work either

Traceback:

Traceback (most recent call last):
  File "/home/developer/PycharmProjects/pandas/pandas/core/groupby/groupby.py", line 2594, in _get_cythonized_result
    vals, inferences = pre_processing(vals)
  File "/home/developer/PycharmProjects/pandas/pandas/core/groupby/groupby.py", line 1368, in objs_to_bool
    vals = vals.astype(bool)
  File "/home/developer/PycharmProjects/pandas/pandas/core/arrays/boolean.py", line 388, in astype
    raise ValueError("cannot convert float NaN to bool")
ValueError: cannot convert float NaN to bool

[arw2019]

For GroupBy that would probably be the best. But we have to fix #37505 before doing this, because this raises for

s = pd.Series([False, pd.NA, False], dtype="boolean", index=[0,0, 1])
grouped = s.groupby(level=0)
print(grouped.any(skipna=False))

too before even reaching the cython part. But I meant the regular DataFrame.any() without groupby. This does not work either

Traceback:

Traceback (most recent call last):
  File "/home/developer/PycharmProjects/pandas/pandas/core/groupby/groupby.py", line 2594, in _get_cythonized_result
    vals, inferences = pre_processing(vals)
  File "/home/developer/PycharmProjects/pandas/pandas/core/groupby/groupby.py", line 1368, in objs_to_bool
    vals = vals.astype(bool)
  File "/home/developer/PycharmProjects/pandas/pandas/core/arrays/boolean.py", line 388, in astype
    raise ValueError("cannot convert float NaN to bool")
ValueError: cannot convert float NaN to bool
``

Ah but "boolean" is an Extension array right? I think if we rewrite in terms of a mask that's a fix and it's preferred performance-wise. But in the meantime we can cast to numpy before passing to cython as was done for std in #37433