feat(entity-manager): pre-computed duplicate detection for large datasets

Replace the live self-join scan (O(N^2), times out at 100K+ authors) with a
background-computed candidates table served instantly at read time.

Key changes:
- Add entity_duplicate_candidates and entity_duplicate_scan_status tables
- Background job batches 1K authors at a time (~58s total for 200K authors)
- Scan endpoint reads from the indexed table (<10ms regardless of scale)
- Candidates are invalidated after merge/delete, not recomputed from scratch
- Add GET /duplicates/status and POST /duplicates/refresh endpoints
- Frontend updates clusters locally after merge/dismiss (no re-scan)
- Status bar shows last computed time and recompute button with progress
- Author browse in entity manager now shows all authors, not just those with books

packages/types/src/entity-manager.ts

@@ -73,6 +73,17 @@ export interface DuplicateScanResponse {
   pageSize: number;
 }
 
+export type DuplicateScanState = 'idle' | 'computing' | 'done' | 'error';
+
+export interface DuplicateScanStatus {
+  entityType: EntityType;
+  state: DuplicateScanState;
+  computedAt: string | null;
+  totalPairs: number | null;
+  threshold: number | null;
+  progressPct: number | null;
+}
+
 // Operation results
 export interface MergeResult {
   targetId: number | string;