Fix translate.py bug (OpenNMT/OpenNMT-py#1317)

[fdalvi]

translate.py would ignore all sentences after processing shard_size*shard_size sentences in src, if no tgt file was provided. This commit fixes this issue (#1317).

I've run all the tests as well, but none of them seem to actually target translate.py. I tested it manually using the following parameters on a ~1000 line source file:

python  translate.py --model model_step_10000.pt --src 1000-line-test.txt --batch_size 1 --shard_size 1 --output pred1.txt
python  translate.py --model model_step_10000.pt --src 1000-line-test.txt --batch_size 1 --output pred2.txt
python  translate.py --model model_step_10000.pt --src 1000-line-test.txt --output pred3.txt
python  translate.py --model model_step_10000.pt --src 1000-line-test.txt --batch_size 1 --shard_size 10 --output pred4.txt
python  translate.py --model model_step_10000.pt --src 1000-line-test.txt --batch_size 10 --shard_size 1 --output pred5.txt

All outputs were identical. If you'd like me to run more tests, let me know!