Enable cross validation/ loso for mlp about nkululeko

[felixbur]

I started on this by introducing the modelrunner class that is used by the runmanager and would encapsulate the cross validation

[felixbur]

i suggest to use a 'fold' column inside the database to designate samples to folds.

[felixbur]

the main problem i see is, to select the best performing model afterwards. Based on which fold?

https://datascience.stackexchange.com/questions/74351/what-is-the-proper-way-to-use-early-stopping-with-cross-validation

[felixbur]

i started doing the changes but skip this for now, there were too many side dependencies on functions i would have to change, based on complexities, e.g. to load models later only for inference without training and such.
I think this would require a proper re-write based on an architecture that models a training for runs, epochs and folds.
Still, fold information that comes with the database should be used with the sklearn based models.