chore: prepare Korean BullshitBench for public release

[wo0jun]

Summary

• reposition ko_native_v0_1 as the primary benchmark and translated_v2 as a comparison baseline
• document the methodology, 0-2 evaluation protocol, control-question policy, and mapping to the original BullshitBench taxonomy
• publish the 48 native verification sources as an ID-linked sidecar
• add contribution, citation, changelog, and machine-readable benchmark metadata
• strengthen dataset validation and run it automatically in GitHub Actions

Why

The repository still described most native rows as placeholders even though all 48 questions were complete. It also lacked the documentation and automated release checks needed for public use. This change brings the repository description, metadata, and validation contract in line with the current dataset.

Impact

• benchmark questions are unchanged
• ko_native_v0_1 is documented as 48 complete questions across all 6 domain by 8 technique combinations
• verification sources remain outside model and judge inputs
• current scoring remains a 0-2 nonsense-detection rubric, with future control accuracy reported separately

Validation

• python3 scripts/validate.py
• python3 -m py_compile scripts/build_release.py scripts/validate.py
• YAML parsing for benchmark.yaml, CITATION.cff, and the validation workflow
• JSON parsing for every release JSONL file
• git diff --check HEAD^ HEAD