fix(audio,sessions): unify audio playback state and improve session discovery

- move voice/audio players to single global audio engine with shared volume

- stop/reset previous track when switching to another media

- keep playback alive across chat switches via global audio element

- list refresh sessions by redis scan fallback when user session set is missing

app/auth/token_store.py

@@ -105,6 +105,26 @@ async def list_refresh_sessions_for_user(*, user_id: int) -> list[RefreshSession
     try:
         redis = get_redis_client()
         session_ids = await redis.smembers(f"auth:user_refresh:{user_id}")
+        if not session_ids:
+            cursor = 0
+            discovered: list[str] = []
+            while True:
+                cursor, keys = await redis.scan(cursor=cursor, match="auth:refresh:*", count=200)
+                for raw_key in keys:
+                    key = raw_key.decode("utf-8") if isinstance(raw_key, bytes) else str(raw_key)
+                    if not key.startswith("auth:refresh:"):
+                        continue
+                    jti = key.removeprefix("auth:refresh:")
+                    if not jti:
+                        continue
+                    owner = await redis.get(key)
+                    if owner and str(owner).isdigit() and int(owner) == user_id:
+                        discovered.append(jti)
+                if cursor == 0:
+                    break
+            if discovered:
+                await redis.sadd(f"auth:user_refresh:{user_id}", *discovered)
+                session_ids = {item.encode("utf-8") for item in discovered}
         sessions: list[RefreshSession] = []
         stale: list[str] = []
         for raw_jti in session_ids: