seq2seq encoder_decoer model, soft and hard attention, batched beam search, attention heat map

1. replace "./en_vi/xxx" with your data files
2. replace "<unk>" and "</s>" with your preferred characters
3. model supports '<space>' splited and no '<space>' splited sentences, so please choose the right value for "FLAGS.whitespace_or_nonws_slip", True means '<space>' splited, False means no '<space>' splited
4. "FLAGS.fix_or_variable_length" define fixed or variable length mode.