seq2seq(attention) have wrong comment

```
context = tf.matmul(attn_weights, enc_outputs)
dec_output = tf.squeeze(dec_output, 0) # [1, n_step]
context = tf.squeeze(context, 1) # [1, n_hidden]
```

I think dec_output shape is [1,n_hidden]

Start with the seq2seq attention example containing the shown tf.matmul and tf.squeeze calls. Verify the shape represented by dec_output, then update its inline comment to match the actual tensor shape; done means the comment is accurate and the surrounding context comment remains consistent.

索引モデルが issue の本文から書いたものです。