Fix MACE fitting

[JaGeo]

Unfortunately, the MACE fitting procedure was broken in the previous version. It only considered energies and forces. See #469 for more details on one of the issues.

This PR should fix the issue. As we haven't produced any production data with this fitting module, this is not an issue for us.

[JaGeo]

I would like to merge this fast. However, my feeling is that it would be good to, in general, test the fitting modules systematically.

Common use cases:

• use a previous database split in train/test, perform fit
• use a previous database already split into train/test, perform fit
• additional data clearning or selection needs to be performed.

[JaGeo]

For now, I will continue in #483 . This PR attempted to much by also restructuring the fitting module imports.