Add Evidence Lab for claim-level RAG reliability

[YuCPbit]

Summary

• add evidence-backed document extraction with exact source offsets and risk-aware review routing
• add a claim-level reliability evaluator for RAG and AI-agent outputs
• verify quote integrity and preserve SHA-256 source provenance
• classify claims as direct, inferred, unsupported, or contradicted
• detect material amount, date, percentage, and negation mismatches
• add an interactive Evidence Lab and a reproducible 10-case gold-set benchmark

Validation

• python -m unittest discover -s tests -v — 14 tests passed
• bash -n scripts/demo-evidenceops.sh
• python scripts/benchmark-evidence-evaluation.py — 10/10 gold-set cases classified correctly
• git diff --check

Portfolio focus

This extension demonstrates a focused AI reliability workflow for contracts, policies, research records, and other source-grounded systems where citations, provenance, contradictions, and abstention decisions matter.