149: Initial changes to unify normalizations and transformations

This MR unifies transformations and normalisation. The main changes are:

• All normalisations are now implemented as a DataObjectTransformation
• DataBlockNormalization is now called DataBlockTransformation - the default class DefaultDatablockTransformation applies the DataObjectTransformation on the DataObject's
• The model handles the one-hot encoding. There is a new layer OneHotEncoding and postprocessors (e.g. LogitPostprocessor, SigmoidPostprosessor, etc.) and new heads: CategoricalInHead, CategoricalOutHead
• The method predict can now be used to do a forward pass, and inverse pass, or an entire model pass. See also the shortcuts generate, predict_y
• The plotter is adapted to the changes, but not all plots work for categoricals yet.

Things left to do:

• Make it possible to pass custom evaluation losses/metrics to the evaluation method
• Adapt the generator
• Adapt all the examples
• Testing for bugs
• Adapt the user guide --> will be resolved later with #154 (closed)

Here an example notebook to test the categoricals test.ipynb