Calculate loss at beginning and end of training

I talked to some users who were a bit confused / surprised that the validation loss didn't get calculated during a run. I think that's a valid issue if you have a small dataset where your training steps are smaller than the default eval.interval=100.

I think the best way to address this is to also always calculate the validation loss always at the beginning and end of each training run. These would also be useful reference points to have in general.

Should be easy for me to add.