fix: parallel subagent cancellation + register task before lock

- cancel_by_session: use asyncio.gather for parallel cancellation
  instead of sequential await per task
- _dispatch: register in _active_tasks before acquiring lock so /stop
  can find queued tasks (synced from #1179)

nanobot/agent/loop.py

@@ -304,9 +304,14 @@ class AgentLoop:
                 ))
 
     async def _dispatch(self, msg: InboundMessage) -> None:
-        """Dispatch a message for processing under the global lock."""
-        async with self._processing_lock:
+        """Dispatch a message for processing under the global lock.
+
+        The task is registered in _active_tasks *before* acquiring the lock
+        so that /stop can find (and cancel) tasks that are still queued.
+        """
         self._active_tasks[msg.session_key] = asyncio.current_task()  # type: ignore[arg-type]
+        try:
+            async with self._processing_lock:
                 try:
                     response = await self._process_message(msg)
                     if response is not None:

nanobot/agent/subagent.py

@@ -268,17 +268,15 @@ When you have completed the task, provide a clear summary of your findings or ac
     async def cancel_by_session(self, session_key: str) -> int:
         """Cancel all subagents spawned under the given session. Returns count cancelled."""
         task_ids = list(self._session_tasks.get(session_key, []))
-        cancelled = 0
+        to_cancel: list[asyncio.Task] = []
         for tid in task_ids:
             t = self._running_tasks.get(tid)
             if t and not t.done():
                 t.cancel()
-                try:
-                    await t
-                except (asyncio.CancelledError, Exception):
-                    pass
-                cancelled += 1
-        return cancelled
+                to_cancel.append(t)
+        if to_cancel:
+            await asyncio.gather(*to_cancel, return_exceptions=True)
+        return len(to_cancel)
 
     def get_running_count(self) -> int:
         """Return the number of currently running subagents."""