docs: Path A — show how to filter or selectively load PR vs IF via class column / TASK_INVENTORY.json

docs: Path B headline is now full-train download+extract, subsets demoted to B.2

docs: add per-task train shard inventory + train-focused Path B example

docs: rewrite 'How to use' for parquet-shipped repo (schema, Path A direct read, Path B extract to dir)

Update README counts (unchanged=46) and note mixed-task percentages

Document train_pr_rewritten and train_pr_unchanged splits

Document Tier-2 extremum-flip variants in OOD/IF

Document prompt_original semantics (presence ≠ modified)

Add train_samples split to configs

Add train split (parquet glob)

Upload README.md with huggingface_hub