OSF backed data

[himjl]

PR Notes
This branch updates hobj’s packaged-data flow to use OSF instead of the old S3 tarball, and makes the cache layout safer for installed-package use.

Key changes relative to v2:

• Switched dataset download in hobj/data/download.py from the S3 .tar.gz to the OSF root zip for node pj6wm.
• Added support for the new archive layout where the outer OSF zip contains data/, and data/images.tar.gz is unpacked into data/images and then deleted.
• Moved the default cache out of the repo root into a versioned user cache path: ~/.hobj_cache/pj6wm-v2/data.
• Added the hobj-download-data console script in pyproject.toml.
• Expanded cache/download coverage in tests/test_data_loader_cache.py.
• Added new tests for dataset-shaped statistics outputs in tests/test_statistics_builders.py and benchmark usage in tests/test_statistics_dataset_usage.py.
• Refreshed the README and example notebooks to reflect the updated data flow and outputs.

Suggested PR summary:
“Replace packaged dataset download flow with OSF-backed versioned cache, add nested image archive extraction, and extend dataset/statistics test coverage.”

Validation run:

• make lint
• make check
• make test