feat: write config to mlflow as artifact

[MartinBernstorff]

• Cross-validation -> Find best model -> New training with same params on train/test split
  • Supported by loading .cfg from cross-validation, getting the relevant modules, and then re-fitting with SplitTrainer
  • ACTION: Write config to disk as .cfg (Implementation outside the trainer? Logger? Disksaver?)

[MartinBernstorff]

Action: Write config to mlflow as artifact (easier reloading)