Make use of concatenation-based data augmentation

[jowagner]

We've implemented per-class concatenation-based data augmentation but haven't used it yet.

• [x] Make a few runs with train.py and predict.py to see can it help ML methods such as DT, SVM and XGBoost.
• [x] Add functionality to save the augmented data in EDOS training data format (construct new rewire IDs and write all expected columns)
• [ ] Try using the augmented data in fine-tuning LLMs