feat: M3 — automatic channel tagging (language + topic system tags)

- Tag and ChannelTag models + migration 0003 (partial unique indexes split
  system vs per-user tag names)
- Offline language detection (py3langid) constrained to a curated language set,
  with the channel's declared default language as a strong prior
- Topic tags mapped from YouTube topicDetails + dominant video category; the
  generic "Lifestyle" catch-all is intentionally dropped
- System (auto) tags are regenerated idempotently and never touch user tags;
  orphaned system tags are cleaned up
- GET /api/tags and admin POST /api/tags/recompute; scheduled autotag job

backend/app/scheduler.py

@@ -6,6 +6,7 @@ from apscheduler.schedulers.background import BackgroundScheduler
 
 from app.config import settings
 from app.db import SessionLocal
+from app.sync.autotag import run_autotag_all
 from app.sync.runner import (
     run_deep_backfill,
     run_enrich,
@@ -48,6 +49,10 @@ def _backfill_job() -> None:
     _job("backfill", work)
 
 
+def _autotag_job() -> None:
+    _job("autotag", lambda db: run_autotag_all(db, only_missing=True))
+
+
 def start_scheduler() -> None:
     global _scheduler
     if not settings.scheduler_enabled or _scheduler is not None:
@@ -65,6 +70,12 @@ def start_scheduler() -> None:
         minutes=settings.backfill_interval_minutes,
         id="backfill",
     )
+    scheduler.add_job(
+        _autotag_job,
+        "interval",
+        minutes=settings.autotag_interval_minutes,
+        id="autotag",
+    )
     scheduler.start()
     _scheduler = scheduler
     logger.info("scheduler started")