Handle zero divs in consistent way

Sklearn has a consistent way of dealing with division-by-zero (either ignore, error out or warn), which we pass on in wrappers (see [example](https://github.com/huggingface/evaluate/blob/main/metrics/precision/precision.py#L143)).

In our own canonical evaluation modules, we also have potential ZeroDivs that we currently do not handle, e.g.: https://github.com/huggingface/evaluate/blob/main/comparisons/mcnemar/mcnemar.py#L94. We should fix this.