fix: dispatch XMPP message callback in thread to avoid blocking asyncio loop

The LLM call (up to 300s) was blocking the slixmpp asyncio event loop,
preventing keepalive pings and causing the server to close the connection.
All XMPP message handling now runs in a daemon thread.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

agents_core/xmpp_client.py

@@ -129,10 +129,19 @@ class XMPPClient:
             logger.debug(f"[XMPP] Message ignoré (non autorisé) : {sender}")
             return
         if self._message_callback:
-            try:
+            # Dispatch dans un thread pour ne pas bloquer la boucle asyncio XMPP
-                self._message_callback(sender, body, is_muc)
+            threading.Thread(
-            except Exception as e:
+                target=self._dispatch_callback,
-                logger.error(f"[XMPP] Erreur callback : {e}")
+                args=(sender, body, is_muc),
+                daemon=True,
+                name="xmpp-handler",
+            ).start()
+
+    def _dispatch_callback(self, sender: str, body: str, is_muc: bool):
+        try:
+            self._message_callback(sender, body, is_muc)
+        except Exception as e:
+            logger.error(f"[XMPP] Erreur callback : {e}")
 
     def send_message(self, to: str, body: str, is_muc: bool = False):
         """Envoie un message XMPP (direct ou MUC)."""