fix(e5): address code-review findings (TMDB guid media_type, search ingest guard, discover keying)

Follow-ups from the epic-end /code-review of the E5 branch:

- tmdb.resolve_recs + people._resolve_known_for: match recommendations/
  credits to the library bucketed by media_type. TMDB movie ids and tv ids
  are independent sequences, so a bare `tmdb:<id>` match could deep-link a
  movie rec to an unrelated show (or vice-versa) on a numeric collision.
  Regression test added (test_resolve_recs_matches_by_media_type_not_bare_id).
- routes/search: wrap the _ingest_candidates call so a DB/enrich failure
  degrades to a 422 with a message instead of a bare 500 hidden behind the
  FE's generic "search failed" (extends the B-core-1 rationale); rollback on
  failure. Stamp enriched_at with its own timestamp taken after the enrich,
  not the stub-insert placeholder.
- routes/plex/discover: key serialized cards back by rating_key (globally
  unique in Plex) instead of id(obj); fix the stale rows 3-tuple annotation.

backend/app/routes/search.py

@@ -12,6 +12,7 @@ Ingested results are flagged (videos.via_search / channels.from_search) so the L
 this search-discovered "noise" by default. Shorts and currently-live/upcoming videos are never
 ingested this way (Shorts are confirmed by the youtube.com/shorts probe; live/upcoming are
 dropped from the search response up front)."""
+import logging
 from concurrent.futures import ThreadPoolExecutor
 from datetime import datetime, timezone
 
@@ -43,6 +44,8 @@ from app.youtube import search_scrape
 from app.youtube.client import YouTubeClient, YouTubeError
 from app.youtube.shorts import make_client, probe_is_short
 
+log = logging.getLogger("siftlode.search")
+
 router = APIRouter(prefix="/api/search", tags=["search"])
 
 # search.list costs this many quota units per page.
@@ -162,12 +165,13 @@ def _ingest_candidates(
             detail_by_id = {it["id"]: it for it in yt.get_videos(ids)}
     except YouTubeError:
         detail_by_id = {}
+    enriched_at = datetime.now(timezone.utc)  # when THIS enrich ran (distinct from the stub placeholder)
     for v in videos:
         item = detail_by_id.get(v.id)
         if item is None:
             continue
         apply_video_details(v, item)
-        v.enriched_at = now
+        v.enriched_at = enriched_at
     db.commit()
 
     # 4) Confirm/deny Shorts (no quota) so none enter via search.
@@ -285,7 +289,19 @@ def search_youtube(
                 and it["channel_id"]
                 and it["channel_id"] not in blocked
             ]
-            collected.extend(_ingest_candidates(db, yt, user, candidates))
+            # Ingest (DB writes + enrich) runs outside the page-fetch try above; guard it too so a
+            # write/shape failure degrades to a 422 with a message on this JSON endpoint instead of a
+            # bare 500 hidden behind the FE's generic "search failed" (same rationale as B-core-1).
+            try:
+                collected.extend(_ingest_candidates(db, yt, user, candidates))
+            except Exception as exc:  # noqa: BLE001 — any ingest failure, not just YouTubeError
+                db.rollback()
+                log.exception("live search ingest failed for %r", term)
+                if collected:
+                    break  # keep what earlier pages already gathered
+                raise HTTPException(
+                    status_code=422, detail=f"YouTube search failed: {exc}"
+                ) from exc
             next_cursor = page["next_page_token"]
             # The API source pages cost 100 units each — never auto-page; one page per request.
             if source == "api" or len(collected) >= limit or not next_cursor: