[Bug] Cross validation yields unexpected additional info

# Issue Description
It seems abstract evaluator does not save cross-validated performance for opt/test datasets.
Also train_evaluator.py is too messy and requires major refactoring.

## Possible Solution
Modify `calculate_auxiliary_losses` in `abstract_evaluator.py`

## Your Code

When using the new results_manager, the following yields the issue.
(Although I fixed it)
```
python -m pytest test/test_api/test_api.py::test_tabular_classification[CrossValTypes.k_fold_cross_validation-resampling_strategy_args1-40981] -W ignore
```