Add dataset catalog, data loaders, and data preparation tooling for versioned datasets - #6

Summary

Introduce a dataset catalog workflow and end-to-end data tooling to register, load, and preprocess versioned datasets for detection and classification tasks, integrated with StreetScenePipeline.

Details

• Implement dataset catalog (src/data/catalog.py) to store versioned datasets with labels, splits, and provenance hashes in data/datasets.yaml
• Add YOLO dataset loader (src/data/yolo_dataset.py) and classification dataset loader (src/data/classification_dataset.py) with preprocessing hooks
• Develop shared augmentation pipelines (src/data/transforms.py) including YOLO mosaics, HSV transforms, and classification policies toggleable via config
• Add scripts/data_prepare.py to ingest raw data, convert to YOLO/classification formats, generate dataset manifests, and update catalog provenance
• Update StreetScenePipeline.prepare_data to resolve dataset entries from the catalog and emit task-appropriate targets
• Document the data lifecycle (acquire → preprocess → register → use) in docs and README
• Extend training/evaluation scripts to support catalog-based datasets while preserving legacy direct-path usage
• Note: No breaking changes introduced

Warning: Task VM test is not passing, cto.new will perform much better if you fix the setup