Add prior-art dedup gate before meta_filter_score LLM call

In runFilterScoreOnOne, before scoreProposal, query engine.db for prior meta-lane hypotheses with same change_type AND any overlap in target_files AND meta_ship_status IN ('proposed','filter_kept','council_kept','shipped') within last 60 days. If a match, short-circuit to a synthetic DROP verdict with reason 'duplicate_of_<prior_id>' and persist via the existing transition wrapper. Also block re-proposing against meta_ship_status='council_rejected' priors within 30 days unless the new proposal's evidence array contains at least one source not present in the prior. No LLM cost on dup hits. New helper findPriorMetaCollisions(metaDesc, db) in lib/db.js for the lookup.

10-25% of incoming meta proposals short-circuit to DROP without LLM call, cutting meta_filter LLM cost proportionally and freeing reviewer attention for novel proposals. Distinct-change_type ratio per cycle rises.

Replay the gate against the last 100 meta proposals already filtered. If fewer than 5% would have triggered, the duplication pattern at the rate claimed does not exist and gate is unjustified at this complexity. Also: if gate triggers on proposals the LLM judge previously KEPT for material novel reasoning, the collision predicate is too broad.

Passed the council verdict. Awaiting Commander decision — approve, reject, or defer.