Using Mlflow and Jaxtrain with --data-generation-experiment-id

Nobody has claimed this yet.

When I run jaxtrain without providing a training data folder and using the experiment ID parameter, it fails with this error:
```
ValueError: Failed to get training data from MLflow and no
--training-data-folder provided: Could not determine training_data_folder from
MLflow experiment 1. Please provide --training-data-folder explicitly.
```

Upon further inspection, I think the error comes from this line in jax_train:
`264: mlflow_data_folder = first_run_info.get("data_output_folder")`

In the `first_run_info` dictionary, I have no key for `'data_output_folder'` even though I do have keys for `'run_id',`` 'run_name'`,` 'num_files'`,` 'total_size_mb'` and `'files'` . It looks like` mlflow_lineage_info` was created, as the logger message `MLflow reports 2 runs with 90 total files` was printed correctly, but the` data_output_folder` is not stored within `first_run_info`

Start in jax_train around line 264 and trace how first_run_info is built from mlflow_lineage_info. Reproduce the --data-generation-experiment-id path without --training-data-folder, then verify that the MLflow-derived training data is resolved without the reported ValueError.

Written by the indexing model from the issue text.