ROUGE evaluation with multi-sentence summaries

[falcondai]

If i understand correctly, the ROUGE functions implemented do support multi-sentence summaries (a summary as a list of strings), i.e., respecting the sentence boundaries when computing n-grams, but the rouge function assumes summaries to be single-sentence (a summary as a string).

I can submit a quick fix on https://github.com/google/seq2seq/blob/master/seq2seq/metrics/rouge.py