EHN: make covariance more flexible

[bruAristimunha]

At the moment, the covariance function from the array api extra does not allow the use of any parameters, such as bias/ddof/fweights/aweights/rowvar.

These numpy and torch parameters make a difference for many applications with covariance. I was wondering if it would be okay to expose and normalize the parameters. It is a really simple PR, a few lines.

[lucascolley]

Could these be added to the agnostic implementation in _funcs.py too?

[bruAristimunha]

Yes, I think so!

[bruAristimunha]

I think it is just a matter of exposing some parameters on the delegation and changing a little the _ funcs.py, I will create to PR ot better discuss with you @lucascolley :)