feat(M1): query endpoint — claims-native, returns ranked_claims with source_id

[ARC345]

Summary

Rebuild the POST /query endpoint to search claims (not papers) and return source_id instead of paper_id.

Changes

• Graphiti search uses group_ids=["claims"] (was ["papers"])
• Response shape: { query, answer, ranked_claims[] } — removes legacy sources list
• RankedClaim has source_id (not paper_id), adds tags
• Context for LLM synthesis is built from claims with confidence scores
• _rank_claims_from_graph uses claims table (not papers)
• Relevance is still a position-based stub (TODO: join on graphiti_node_uuid for real per-claim scores)

Files

• backend/api/query.py — rewrite
• backend/tests/test_query.py — rewrite