Add reproducible prior art evidence matrix

[GiulioDER]

Summary

This pull request adds a reproducible, reviewable prior art evidence system for RE-call.

Included

• Structured JSON and JSON Lines records for sources, systems, taxonomy, claims, and reviews
• Deterministic capability matrix, combination analysis, gap report, and machine readable summary
• Fail closed validation for references, evidence, dates, source tiers, reviews, and target capabilities
• Primary evidence expansion covering agent memory, rollback, action feedback, deletion, provenance, and memory security evaluation
• CI validation for canonical records, generated reports, and prior work declarations in changed experiments
• Focused tests for schemas, validation, rendering, determinism, uncertainty states, and experiment declarations

Evidence boundary

The reports preserve verified, partial, not_evidenced, contradicted, and unknown separately. They identify research candidates and do not automatically claim that RE-call is first, unique, or unprecedented.

Verification

• python -m tools.prior_art validate
• python -m tools.prior_art render --check
• Ruff checks passed
• 16 focused prior art tests passed
• 76 published artifact tests passed

No production retrieval or memory runtime behavior changes in this work.