Fix a bug in the last decoding step of training CopyNet

[GregoryZeng]

Closes allenai/allennlp#5354

[epwalsh]

Thanks @GregoryZeng! This is great. I think the comment on lines 495-496 is still confusing:

        # The last input from the target is either padding or the end symbol.
        # Either way, we don't have to process it.
        num_decoding_steps = target_sequence_length - 1

I think it would be more clear if it said something like:

        # We have a decoding step for every target token except the "START" token.
        num_decoding_steps = target_sequence_length - 1

Do you agree? If so, could you update that in this PR as well?

[GregoryZeng]

Sure, that's definitely clearer.

[epwalsh]

Looks great! Thanks @GregoryZeng!