feat: persist duplicate detection results in PostgreSQL

[JohnRDOrazio]

Problem

Duplicate detection results are currently cached only in Redis with a 10-minute TTL. When the page is reloaded, results are gone and the user must click "Find Duplicates" again (~56 seconds for a 15K-class ontology).

Proposal

1. New duplicate_detection_results table — stores the latest detection results per project/branch, including clusters, threshold, and timestamp
2. Auto-update on index rebuild — when run_ontology_index_task completes (or the index is updated via entity edits), automatically re-run duplicate detection and persist the results
3. Frontend loads persisted results on mount — the Duplicates tab should check for stored results first, showing them immediately without requiring a manual "Find Duplicates" click
4. "Find Duplicates" button re-runs detection — still available for on-demand refresh, updating the persisted results

Context

PR #80 moved duplicate detection to the ARQ worker queue and rewrote it to use PostgreSQL's pg_trgm GIN index instead of in-memory rdflib parsing. The detection itself now completes in ~56 seconds for large ontologies. Persisting results would make the UX seamless.

Tasks

• Create duplicate_detection_results table and Alembic migration
• Store results after run_duplicate_detection_task completes
• Load persisted results in the Duplicates tab on mount (frontend)
• Trigger duplicate detection automatically after run_ontology_index_task
• Update "Find Duplicates" to refresh persisted results