Add encoder for 'requested slots' classification

-Implementing the encoder and dialogue representation states using LSTM in 'model.py' -Trying classifications for requested slots (binary values). It's working and reaching ~92% on dev set. The model and training setup follow the paper (not identically, as close as possible)

-Adding one line in Simon's code to get the kvret vocabulary. Because using data.train.vocabulary._size doesn't seem to work