fix(multimodal): harden playback and client lifecycle handling

- retry the server connection prompt after transient playback failures
- serialize multimodal client and session start-stop operations
- tag downstream responses to ignore stale packets and playback events
- complete bounded teardown and clean callbacks after partial startup

components/control_plane/include/control_plane.h

@@ -154,8 +154,9 @@ esp_err_t control_plane_start_multimodal_client(void);
  * Semantics:
  * - Idempotent and safe to call repeatedly.
  * - Requests cooperative stop first, then applies bounded waits for worker teardown.
+ * - Returns an explicit error if teardown does not complete.
  */
-void control_plane_stop_multimodal_client(void);
+esp_err_t control_plane_stop_multimodal_client(void);
 
 esp_err_t control_plane_register_model_command_callback(control_plane_model_command_callback_t cb,
                                                         void *user_data);