Add fixed point covariance estimator and add **kwds arguments in Covariances

[qbarthelemy]

This PR:

• adds a new covariance estimator: fixed point covariance matrix;
• adds **kwds arguments in Covariances, for parameters passed directly to the covariance estimator.

[sylvchev]

Nice addition! You could add the whatsnew and it is good to merge.

[qbarthelemy]

Yep, but **kwds arguments must also be added
in functions: covariances_EP, covariances_X, block_covariances;
and in classes: ERPCovariances, XdawnCovariances, BlockCovariances, HankelCovariances.

[agramfort]

I find the name not super informative and reading the PR it's not super clear what it is useful for and why I should use it? Can you add a blob of text somewhere to explain this?

🙏

[agramfort]

What am I missing here @qbarthelemy ? See the estimate is very different from Id matrix.

[qbarthelemy]

Good catch @agramfort !

I had naively translated code from Matlab covariance toolbox https://github.com/alexandrebarachant/covariancetoolbox/blob/master/lib/estimation/FPCov.m
where trace-normalization is missing (Eq(8) in Pascal2005).

It is corrected and I have completed tests.

[agramfort]

victory !

Feel free to just add a line in what's new saying that it's a robust estimator.

+1 for MRG

thx @qbarthelemy

[sylvchev]

Thanks @qbarthelemy for the PR and thank you @agramfort for spotting this!