Rework Machine Learning pipeline

[mafrahm]

This PR:

• adds a task for input preparation before the training (hbw.MLPreTraining)
• adds a task for evaluation of a single fold using the MLPreTraining output (hbw.MLEvaluationSingleFold)
• adds a task for plotting using outputs from all previous tasks (hbw.PlotMLResultsSingleFold)

We also added two classes for ML Dataset preparation; one class preparing the output arrays for hbw.MLPreTraining, and one class that conveniently loads these outputs from hbw.MLPreTraining back into memory.

Next steps:

• we could produce SHAP values in the hbw.MLEvaluationSingleFold task aswell, since calculating them takes quite long
• bookkeeping of parameters can be improved
• improve storing of introspection results
• more plots?