Add eval_batch_size

[cosmaadrian]

Update the config to use an eval_batch_size (usually larger than train_batch_size.)

Update the evaluate.py to be able to receive a custom eval_batch_size if given (it currently uses the eval_batch_size from the loaded config.)