feat: Phase F F10 negative-result citation policy

[cschanhniem]

Summary

• Adds NRC- negative-result citation policy schema (src/openamp_foundry/evidence/negative_result_citation_policy.py)
• Machine-enforces that every public positive claim cites all known relevant NRR- negative results
• Three-verdict system: compliant / non_compliant / no_relevant_negatives
• 50 tests; BASELINE updated 10924→10974; F10 marked complete in NEXT_100_PR_MAP.md

Design

• relevant_nrr_ids: list of NRR- records that are relevant to the positive claim being made
• cited_nrr_ids: list of NRR- records actually cited/referenced in the claim
• uncited_nrr_ids, uncited_count, all_relevant_cited, policy_verdict all auto-computed
• A claim with uncited negatives is flagged non_compliant — no silent omission possible
• cited_nrr_ids must be a subset of relevant_nrr_ids (no citing irrelevant records to pad count)
• dry_lab_only=True always enforced

Safety

• Directly addresses cherry-picking: the most common way an AI pipeline overstates its case
• Does not weaken any existing safeguard
• Completes Phase F (all F1–F10 items now done)
• Marks F10 complete in NEXT_100_PR_MAP.md

Test plan

• 50 unit tests pass (constants, build, validate, format)
• BASELINE regression test passes (10974)