Improved stage 2 training pipeline

Major change:

In stage 2, model is now trained on one half of the mixed dataset and validated on the other, without splitting off validation data in each half. This is more exhaustive and logical.

Minor change:

Scripts are renamed to streamline the workflow

Data:

Three models are provided that allow to run notebook #03