Poincare: Add generic getFeatureInfo method to dataset class

[egillax]

The initial layers in the models need some info about the data to be initialized with the correct tensor sizes.

Currently this is handled by two methods here get_cat_features and get_num_features. Methods defined here

Now when we are adding a new type of features, embeddings that have already been generated, we should rethink this approach.

I think we should add a new method to the dataset class, something like get_feature_info to replace the other two. The output of that would be stored in the modelParameters$featureInfo. This way the dataset can provide all the info needed and the model needs to deal with extracting the correct bits to initalize it's layers correctly.