fix(python): fix ufunc in agg (change __ufunc_array__ so it uses is_elementwise=True parameter)

[deanm0000]

Fixes #13557

I was back burnering whether or not there was a good way to tell if a ufunc was elementwise for a while now when I realized that they're all elementwise.

I did this test:

import numpy as np
import polars as pl
import scipy.special as ss
from polars.testing import assert_frame_equal
npufuncs = []
for x in dir(np):
    if getattr(np,x).__class__==np.ufunc:
        npufuncs.append(x)
ssufuncs = []
for x in dir(ss):
    if getattr(ss, x).__class__==np.ufunc:
        ssufuncs.append(x)
        
df=pl.DataFrame({'a':[1,1,2,2],
                 'b':[1.1,2.2,3.3,4.4],
                 'c':[3.1,4.2,5.3,6.4]})

goodufuncs = {"<module 'numpy' from '/usr/local/lib/python3.11/site-packages/numpy/__init__.py'>":[],
              "<module 'scipy.special' from '/usr/local/lib/python3.11/site-packages/scipy/special/__init__.py'>":[]}
for cls, xlist in {np:npufuncs, ss:ssufuncs}.items():
    for x in xlist:
        onearg=True
        while True:
            try:
                if onearg:
                    assert_frame_equal(
                        df.group_by('a',maintain_order=True).agg(pl.col('b').map_batches(getattr(cls,x), is_elementwise=True)),
                        df.group_by('a',maintain_order=True).agg(pl.col('b').map_batches(getattr(cls,x), is_elementwise=False))
                    )
                else:
                    assert_frame_equal(
                        df.group_by('a',maintain_order=True).agg(pl.struct('b','c')
                            .map_batches(lambda z, cls=cls, x=x: getattr(cls,x)(z.struct.field('b'), z.struct.field('c')), is_elementwise=True)),
                        df.group_by('a',maintain_order=True).agg(pl.struct('b','c')
                            .map_batches(lambda z, cls=cls, x=x: getattr(cls,x)(z.struct.field('b'), z.struct.field('c')), is_elementwise=False))
                    )
                goodufuncs[str(cls)].append(x)
                break

            except Exception as err:
                if "takes from 2 to " in str(err):
                    onearg=False
                else:
                    print(f"{x} gave {err}")
                    break

At the end, there are no assertion errors so for stock ufuncs in numpy and scipy.special it is safe to dispatch them with map_batches(func, is_elementwise=True) For any other ufunc, like the result of using numba.guvectorize there will be a warning that they should call their function through map_batches.

[ritchie46]

Ah, ok, thanks for testing that. 😄 Thanks for the this change. It should probably to less surprises.