Docs of dimensions of decoder RNN are wrong

[kylegao91]

• inputs should have size (batch_size, seq_len, hidden_size) since RNN is set to be batch first.
• encoder_hidden should be (num_layers * num_direction, batch_size, hidden_size)
• decoder_outputs should be (seq_len, batch_size, vocab_size)

[yucoian]

I also think the docs are easily misleading in decoderRNN especially. I suggest add more description on the decoderRNN class's member function , its inputs and outputs. The rest docs are really nice.