Fixed a bug and cleaned up the code in continue_from section of train.py

[SiddGururani]

Rebased my fork of the repo with the new changes and tested out some logging features. There appears to have been a bug in this line:

loss_results, cer_results, wer_results = package['loss_results'], package['cer_results'], package[
                'wer_results']

The variables loss_results, etc are all of length args.epoch and so tensorboard logs had the full tensor logged instead of only the values up until start_epoch. My bad, because I was the one who wrote that piece of code. I think I fixed it at some point and forgot to submit another pull request.

[SeanNaren]

thanks for taking the time to fix this!