Add `zero_division` to F1 metric

[TimRepke]

The underlying sklearn function exposes the zero_division attribute [1], which is not a keyword argument in the metric wrapper in this library. This means, users get a warning when there is a zero division with the note to set this argument to prevent the warning, which, when set, causes an error. This issue was described several years ago [2] and still persists in the latest version.

This PR should fix that.

[1] https://scikit-learn.org/stable/modules/generated/sklearn.metrics.f1_score.html [2] https://discuss.huggingface.co/t/zero-division-warning-in-metric-compute/21386