Refactor data retrieval, model I/O, and training logic for improved modularity and flexibility

- Added  to support organized data downloads by robot UUID.
- Enhanced  and  functions with typing and documentation.
- Introduced functions for saving and loading configurations to/from JSON files.
- Updated training workflow to handle pre-trained model layers: selectively remove last layers and freeze them before adding new layers.
- Implemented flexible setup for W&B sweeps, including hyperparameter configuration from external data.
- Streamlined upload process by refactoring config saving logic and reducing redundant code.
- Improved logging for data analysis and sweep setup for better traceability.