[FIX] Handle polars string columns in the missing-value check

[Raashish Aggarwal (raashish1601)]

No existing issue.

Fitting on a polars frame that has a string categorical column (declared in cat_exog_list) fails before the vocabulary is even built:

polars.exceptions.InvalidOperationError: `is_nan` operation not supported for dtype `str`

_check_nan runs ufp.is_nan_or_none on every temporal and static column. For polars that calls Series.is_nan(), which only works on float columns. pandas is fine because it uses isna(). The categorical encoding already has a polars branch, so this check was the only thing blocking it.

The check now uses is_null() for non-float polars columns and keeps the old path for everything else. Nulls in a string column are still reported as missing values.

Added test_polars_string_categoricals_run in tests/test_categorical.py. It fits and predicts with hist, futr and static string categoricals on polars, and checks that a null in a string column still raises. It fails on main and passes with this change. tests/test_categorical.py passes locally, and so do the mask/missing-value tests in tests/test_core.py.

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[Marco (marcopeix)]

This fix should be in utilsforecast, because the issue is really from the function is_nan_or_none.

[Raashish Aggarwal (raashish1601)]

Makes sense, is_nan_or_none is the right place. I'll open the fix in utilsforecast and close this PR once that is in.

[Raashish Aggarwal (raashish1601)]

Opened the fix in utilsforecast: Nixtla/utilsforecast#297. is_nan_or_none now uses is_null() for non-float polars columns.

[Raashish Aggarwal (raashish1601)]

Closing in favor of Nixtla/utilsforecast#297, which fixes this in is_nan_or_none.