Missing Eref

[cortner]

If no reference 1-body energies are supplied then the fitting routine should look for single-atom-in-vacuum training data and use those energies. If such data doesn't exist, then a warning should be printed that we highly recommend use of reference energies.

More importantly though - the basis should then expanded to allow for fitting of the one-body terms.

[cortner]

GAP and MACE seem to do it via a kwarg E0s = "isolated" where "isolated" is the configtype of the structures that contain the E0s.