Adjusting DenseForSparse layer size

[jupyterjazz]

After completing #8749 and #8750, in order to dynamically change sparse layers during fine-tuning, we will need to adjust the DenseForSparse layer size in case there are any new sparse features.

This functionality will be implemented in the DIETclass. We will compare persisted feature_sizes to the current feature_sizes to find out if there are any DenseForSparse layers to be updated. After updating the layers, we will compile the whole model, update data signatures, create a RasaModelData object from a data example and activate new layers by fitting the model on the RasaModelData object.

Things to be done:

• call adjust_layers from DIETClassifier during fine-tuning
• find out which DenseForSparse layers should be updated
• adjust each of them
• compile the model
• update data signatures
• activate new layers

[tttthomasssss]

@ka-bu assigned as reviewer.