[ENH, API] PytorchEstimator verbosity about dlordinal

[victormvy]

To show the loss for each epoch, the _fit() method should return the loss for the batch, so the average loss for each epoch can be computed in the fit() method.

[victormvy]

It seems like a good opportunity to add callbacks to the estimator, so that, for example, any metric in addition to the loss value can be calculated and printed. This would replace the verbose parameter.

What do you think?

[franberchez]

This issue is closed because it is related to issue #66