BenchClaw Harness v0.2.0

Citation-ready release of the BenchClaw reproducible benchmark evidence layer.

Highlights:

• Adds Apache-2.0 license and CITATION.cff metadata.
• Publishes the sanitized 160-run LangGraph vs Pydantic AI evidence bundle.
• Adds direct results, manifest, and reproduction links.
• Preserves the verified test-path fixes used by the public harness.

Verification before release:

• 10/10 unit tests passed.
• Bundle checksums and manifests validated.
• 160-record redaction scan passed.