fix(curate): emit valid LeRobot v3 dataset layout

[rylinjames]

Summary

• Change the LeRobot v3 converter from v2-style per-episode parquet files plus JSONL metadata to v3 file-based shards: data/chunk-000/file-000.parquet and meta/episodes/chunk-000/file-000.parquet.
• Write meta/tasks.parquet with pandas-compatible task index metadata and meta/stats.json with action/state statistics.
• Populate info.json totals and v3 path templates, and enforce float32 action/state parquet schemas to match declared features.
• Add regression coverage that rejects the old episode_*.parquet/tasks.jsonl/episodes.jsonl layout and verifies converter output passes the dataset validator.

Validation

• PYTHONPATH=$PWD/src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m pytest tests/test_curate_format_converters.py tests/test_dataset_validator.py -p no:cacheprovider
• PYTHONPATH=$PWD/src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/ruff check src/tether/curate/format_converters/lerobot_v3.py tests/test_curate_format_converters.py
• PYTHONPATH=$PWD/src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m py_compile src/tether/curate/format_converters/lerobot_v3.py tests/test_curate_format_converters.py

🤖 Generated with Claude Code