Minor Fix - Final Batch Size if Total Samples to use in FID Calculation is a multiple of Batch Size

[tempoxylophone]

Thank you so much for open sourcing this work! I think I may have found a very, very slight issue.

When using the --eval-total-size argument, the final batch in the evaluator will be 0 if the value passed to --eval-total-size is a multiple of the batch size, i.e. self.eval_total_size % self.eval_batch_size = 0.

python train.py --eval --eval-total-size 512

This will cause a ValueError later in the FID calculation since we compute the mean of an empty array.