[PIR][WP27] CAMA-pattern correlation-aware memory arbitration — effective evidence lineages (ADR-330)

Part of PIR Wave 4 — epic #837.

Implement arbitration.rs in crates/ruvector-agent-memory: cluster retrieved memories by causal ancestry via fusion.rs::resolve_provenance(), count effective independent evidence groups, confidence = sum over lineages of (source_reliability x freshness), strictly downgrade-only vs naive count, non-finite inputs rejected at the choke point (Wave-3 #888 lesson). Dedicated test: ten memories from one signed source => exactly one effective lineage (acceptance-test clause 3, verbatim). From-scratch — no CAMA (arXiv:2608.19701) code exists; abstract has zero quantitative results.

• Evidence grade & artifact availability: docs/research/perpetual-intelligence-runtime/10-wave4-evidence-review.md (PR #892)
• WP/ADR mapping, phases, and security gates: docs/research/perpetual-intelligence-runtime/11-wave4-program-plan.md (PR #892)
• ADR: docs/adr/ADR-330-cama-pattern-correlation-aware-memory-arbitration.md

Depends on: WP4 (ledger/observation), WP18 (fusion), WP22 (diagnostic-gate registration).

Start with docs/adr/ADR-330-cama-pattern-correlation-aware-memory-arbitration.md and the wave-4 program plan, then inspect crates/ruvector-agent-memory/fusion.rs::resolve_provenance() and the dependencies on WP4, WP18, and WP22. The work is complete when arbitration.rs groups memories by causal ancestry, counts independent lineages, applies the specified downgrade-only confidence rule, rejects non-finite inputs, and passes the clause-3 test for ten memories from one signed source.