fix: report short series in cross_validation for polars inputs

[Moury Bachina (MouryB)]

Fixes #1272

Summary

When cross_validation finds series that are too short for the requested windows, it builds the error message with self.uids[short_series]. That boolean-mask __getitem__ works for pandas but raises TypeError: selecting rows by passing a boolean mask to __getitem__ is not supported for polars, which hides the intended ValueError.

The mask is now applied with utilsforecast.processing.filter_with_mask, which core.py already uses (via ufp) to handle pandas and polars the same way. Behavior for pandas is unchanged.

Tests

Added test_cross_validation_short_series_message, parametrized over pandas and polars, with two short series so the message lists more than one id. The polars case fails on the current main and passes with this change; the pandas case passes on both.

uv run pytest tests/test_core.py -q          # 50 passed
uv run pre-commit run --files python/statsforecast/core.py tests/test_core.py   # ruff, mypy passed

Note

This PR was written with an AI coding assistant (Claude Code).

[CLAassistant]

All committers have signed the CLA.