Add a threshold for the validation chi2

Created by: scarlehoff

Every now and then you can hit a plateau where the validation chi2 does not improve for some number of epochs, then the fit finishes early with a really bad chi2. This threshold (which can be added to the fitting parameters) will force n3fit not to consider the run as acceptable until the threshold has not been reached.

This feature was in the first versions of n3fit (when the alpha and beta exponents were fitted) and the fit rarely stopped too early so I removed it (I think it went away with the refactoring of the stopping). Now that most of the runs are done with trainable: False this seems to be more of a problem. I imagine the fixed exponents can give the NN a hard time at times.

The documentation of this new feature of the runcard will be in #767