Add Seq2Seq Encoder-Decoder GRU with Luong Attention trained on Multi30K

initial commit