SIENTIAPDE-1251: Implement ML model training activity and repository

This commit introduces the 'Training' activity and 'TrainingRepository' for handling ML model training operations within the Model Manager system.

- Added model_manager/activities/training.py for the Training activity, which extends BaseActivity and integrates with Temporal workflows.
- Added model_manager/utils/repository/training_repository.py for the TrainingRepository, which encapsulates the core training logic.
- Updated model_manager/activities/activities.py to include the Training activity in the main activities orchestrator.
- Updated README.md to document the new 'Training' component.
- Added unit tests for the new activity and repository.

README.md

@@ -154,6 +154,12 @@ The Model Manager system uses a Temporal-based workflow architecture with clear
   - Support for three update types: STATUS, STATUS_WITH_ERROR, MODEL_SAVED
   - Automatic error message truncation (1024 chars)
   - Connection pooling and retry logic via Postgres base class
+- **Training**: ML model training operations (standalone activity, composition pattern)
+  - Unified `train_model()` method for complete training pipeline
+  - Receives pre-downloaded files (BytesIO) to avoid memory leaks
+  - Returns success/failure status with TrainModelResult or error message
+  - No exception raising on failure - allows workflow to handle errors gracefully
+  - Integration with TrainingRepository for business logic separation
 - **Gates**: Data quality validation and filtering mechanisms
 - **MLFlow**: Model transformation and prediction operations
 - **MinIO**: Object storage operations for file management