[Convex mean] try graphical_lasso

[gcattan]

Learning from POC: #132

We can try to replace Shrinkage by GraphicalLasso for regularization of the empirical covariance (inside mean.py)
https://scikit-learn.org/stable/modules/generated/sklearn.covariance.graphical_lasso.html#sklearn.covariance.graphical_lasso

[gcattan]

Note, the main issue is that when using convex mean in association with a log-defined metric for distance, the logarithm of the matrices contains sometimes NaN values.

[gcattan]

Investigated within #130.
Graphical_lasso does not provide good results.