fix: send system idle instead of tts stop to avoid Speaking→Listening race

Root cause: tts stop triggers Speaking→Listening, which calls
SendStartListening in HandleStateChangedEvent. idle arrives too late
and gets overridden. Fix: skip tts stop when idle is needed, go
Speaking→Idle directly.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/main/xiaozhi-server/core/handle/sendAudioHandle.py

@@ -44,15 +44,18 @@ async def sendAudioMessage(conn: "ConnectionHandler", sentenceType, audios, text
 
     # 发送结束消息（如果是最后一个文本）
     if sentenceType == SentenceType.LAST:
-        await send_tts_message(conn, "stop", None)
-        conn.client_is_speaking = False
-        # Send system idle command if LLM requested it (e.g. antaf bridge unavailable)
         if getattr(conn, 'send_idle_after_tts', False):
+            # Send system idle instead of tts stop — goes Speaking→Idle directly,
+            # avoids Speaking→Listening which triggers SendStartListening
             conn.send_idle_after_tts = False
+            conn.client_is_speaking = False
             await conn.websocket.send(
                 json.dumps({"type": "system", "command": "idle"})
             )
-            conn.logger.bind(tag=TAG).info("Sent system idle to device")
+            conn.logger.bind(tag=TAG).info("Sent system idle to device (skip tts stop)")
+        else:
+            await send_tts_message(conn, "stop", None)
+            conn.client_is_speaking = False
         if conn.close_after_chat:
             await conn.close()