Add data augmentation for small dataset regime

**Description:**
The dataset has only 140 samples (30-42 per client). Data augmentation could improve model robustness.

**Suggestions:**
- Add Gaussian noise to features
- Implement mixup or cutmix for tabular data
- Add random feature dropout during training

**Files to Modify:**
- `src/data/loader.py` - Add augmentation transforms
- `src/data/preprocessor.py` - Add augmentation pipeline
