Reduce parallel YouTube request workers to avoid cookie invalidation

8 simultaneous yt-dlp processes hitting video pages looks like a bot
attack and causes YouTube to nuke the session cookies. Drop to:
- Popular fetch view_count enrichment: 8→3 workers
- Discovery search: 8→4 workers
- Graph signal (featured channels): 8→3 workers

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/services/discovery.py

@@ -112,7 +112,7 @@ def _search_and_store(
         except Exception:
             return []
 
-    with ThreadPoolExecutor(max_workers=8) as pool:
+    with ThreadPoolExecutor(max_workers=4) as pool:
         futures = {pool.submit(_do_search, q): q for q in queries}
         for fut in as_completed(futures):
             for video in fut.result():
@@ -620,7 +620,7 @@ def update_graph_signal(db: Session, user_id: int):
             return []
 
     featured_map: dict[str, list[str]] = {}
-    with ThreadPoolExecutor(max_workers=8) as pool:
+    with ThreadPoolExecutor(max_workers=3) as pool:
         futures = {pool.submit(_fetch, row["youtube_channel_id"]): row for row in sample}
         for fut in as_completed(futures):
             row = futures[fut]