feat(backend): #356 영상-식당 관련도 LLM 평가

- DB V20260615b: video_restaurants.{relevance, relevance_reason, relevance_evaluated_at} + idx_vr_relevance
- VideoRelevanceService (#322 패턴): @Async verifyAsync + verify + verifyAll(batchSize)
- PipelineService.processExtract → linkVideoRestaurant 후 verifyAsync(linkId) 자동 트리거
- GET /api/restaurants/{id}/videos: 기본 strong/unknown만, ?include_weak=true 시 모두 + relevance/reason
- AdminVideoRelevanceController: GET pending / POST all / POST {id}/evaluate / PATCH {id}
- 캐시 키 strong|all 분리, LLM 실패 시 unknown 안전 기본값(표시 유지)

Refs: #356 (close)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend-java/src/main/resources/db/migration/V20260615b__video_relevance.sql

@@ -0,0 +1,7 @@
+-- #356 영상-식당 관련도 LLM 평가
+ALTER TABLE video_restaurants ADD (
+  relevance VARCHAR2(16) DEFAULT 'unknown' NOT NULL,
+  relevance_reason VARCHAR2(120),
+  relevance_evaluated_at TIMESTAMP
+);
+CREATE INDEX idx_vr_relevance ON video_restaurants(relevance);