Scoring functions

[bellet]

Besides the default score method of our estimators, we should properly test and document the fact that one can use various metrics to evaluate our supervised and weakly-supervised metric learners when using sklearn's tools such as model_selection.GridSearchCV and model_selection.cross_val_score through their scoring parameter.

This should be done after dealing with #131, which will introduce a predict method for weakly supervised pairwise metric learners.