`_log_params()` in `train.py` is not necessarily logging the correct optimizer/criterion and parameters

Nobody has claimed this yet.

The `_log_params()` function in `train.py` currently logs the optimizer and criterion as follows:
```
# Log the criterion and optimizer params
criterion_name = config["criterion"]["name"]
mlflow.log_param("criterion", criterion_name)
if criterion_name in config:
mlflow.log_params(config[criterion_name])

optimizer_name = config["optimizer"]["name"]
mlflow.log_param("optimizer", optimizer_name)
if optimizer_name in config:
mlflow.log_params(config[optimizer_name])
```
The issue is that the user may have overriden the config by directly defining these attributes and/or their parameters in the model, so the config is not necessarily correct. We should try recovering the correct names and parameters from the model instead of just copying what is in the config,

Start in train.py at _log_params() and trace how the model exposes its optimizer, criterion, and their parameters. Compare those values with the current config-based logging, then exercise the training and logging path to confirm MLflow records the model-selected names and parameters.

Written by the indexing model from the issue text.