Improve evaluate (custom loss, ...)

[ppdebreuck]

Improvements such that FitGenetic can handle custom loss:

• add custom loss to genes
• pass it to evaluate

Custom loss functions are only meant for regression. Error is raised if loss is an unrecognised string. ROC AUC still used for classification.

[ppdebreuck]

Ok let's do as such.