fix: support non-float polars columns in is_nan_or_none

[Raashish Aggarwal (raashish1601)]

is_nan_or_none fails on polars columns that aren't numeric, because polars only supports is_nan on numeric dtypes:

import polars as pl
from utilsforecast.processing import is_nan_or_none
is_nan_or_none(pl.Series(["a", None]))
# polars.exceptions.InvalidOperationError: `is_nan` operation not supported for dtype `str`

This shows up in neuralforecast when fitting on a polars frame with a string categorical exogenous column: _check_nan calls is_nan_or_none on every column (see Nixtla/neuralforecast#1634, where it was suggested to fix it here).

For polars series that are not floats, is_nan_or_none now returns is_null(), since only float columns can hold NaN. Float columns and pandas are unchanged.

Extended test_is_nan_or_none with polars string, boolean, integer and categorical series containing a null. It fails on main and passes with this change. tests/test_processing.py passes apart from three polars concat/backtest tests that also fail on main with the polars version I have locally (2.0, above the <=1.31 pin).

[José Morales (jmoralez)]

please move this to the is_nan function. in the polars branch we can check if its float and call is_nan and otherwise return a series filled with false

[Raashish Aggarwal (raashish1601)]

Done in a4c1d45. is_nan now calls is_nan only for polars float columns and returns an all-false series otherwise, and is_nan_or_none is back to is_nan(s) | is_none(s).

[José Morales (jmoralez)]

please move these to the is_nan test

[Raashish Aggarwal (raashish1601)]

Moved them to test_is_nan in a4c1d45. They check is_nan on string, boolean, integer and categorical series, and is_nan_or_none on the same series.