feat: add runtime listen-only switch and API controls

qingjian_cs/app/client.py

@@ -20,6 +20,7 @@ from .logger import setup_logger
 from .observability import activity_event, build_trace_id
 from .orchestrator import Orchestrator
 from .rules import extract_image_urls, prefilter_message
+from .runtime_switch import is_listen_only
 
 
 class QingjianClient:
@@ -214,6 +215,14 @@ class QingjianClient:
         return False
 
     async def _handle_decision(self, data: dict, merged_msg: str, *, auto_quote: bool = False) -> None:
+        if is_listen_only():
+            activity_event(
+                self.logger,
+                "ai_reply_skipped",
+                customer_id=data.get("from_id", "-"),
+                reason="listen_only_mode",
+            )
+            return
         key = self._customer_key(data)
         trace_id = build_trace_id(data.get("acc_id", ""), data.get("from_id", ""), merged_msg)
         t0 = time.perf_counter()
@@ -402,6 +411,15 @@ class QingjianClient:
         key = self._customer_key(patched)
         self._append_dialogue(key, "user", patched["msg"])
 
+        if is_listen_only():
+            activity_event(
+                self.logger,
+                "ai_reply_skipped",
+                customer_id=patched.get("from_id", "-"),
+                reason="listen_only_mode",
+            )
+            return
+
         # 硬编码：每个客户首条消息先快速回复“在的”
         if key not in self.first_msg_replied:
             await self.send_reply(patched, "在的")