Implemented mmd between Gaussian mixture components to track cluster overlapping. Adding a term to the loss function is an option

Implemented version of SEQ_2_SEQ VAE based on tensorflow_probability, and SEQ_2_SEQ_GMVAE draft using tensorflow_probability

Instanciated a mean and a variance for each component, and the categorical prior as a Dense layer with a softmax activation

Restructured models.py and hypermodels.py. Now all helper functions and custom layers are imported from model_utils.py