auto_find_batch_size for OOM during evaluation

### System Info When running training and evaluation (`_inner_training_loop` in the HF `Trainer`), the auto-find-batch size tries to reduce the training batch size, even when the OOM happens during evaluation and not the training. It should distinguish whether the error comes from training or evaluation (`_maybe_log_save_evaluate`) and reduce the evaluation batch size if needed. Also, it would be nice to report the finally chosen batch size (maybe this is already done). ### Who can help? _No response_ ### Information - [ ] The official example scripts - [ ] My own modified scripts ### Tasks - [ ] An officially supported task in the `examples` folder (such as GLUE/SQuAD, ...) - [ ] My own task or dataset (give details below) ### Reproduction Choose a very large evaluation batch size, and a small training batch size. ### Expected behavior Should reduce eval batch size, not just the training batch size