补充 ai_auto_response 的完整模型决策日志

变更项：
1. 新增 LLM_RESULT 日志，记录模型输出的 should_reply、reply_mode、topic、reply 预览和原始响应预览。
2. 新增 BLOCKED_REPLY 日志，记录模型原本想回复但被 post_llm_cooldown、过期、覆盖或重复回复拦下的具体原因。
3. 保留原有 SKIP 与 SENT 日志，使模型判定、发送阻断和最终发出三段链路可以串起来排查。

plugins/ai_auto_response/runtime/logging.py

@@ -76,6 +76,31 @@ def build_log_summary(event: str, data: Dict[str, Any]) -> str:
             f"err={data.get('last_error', '')}"
         ).strip()
 
+    if event == "llm_result":
+        return (
+            f"[XIAONIU] LLM_RESULT room={room} user={sender} "
+            f"trigger={data.get('trigger_type', 'none')} "
+            f"want={yn(data.get('should_reply'))} "
+            f"mode={data.get('reply_mode', '')} "
+            f"topic={data.get('topic', '-') or '-'} "
+            f"reply={data.get('response_preview', '')} "
+            f"raw={data.get('raw_response_preview', '')}"
+        ).strip()
+
+    if event == "blocked_reply":
+        age_text = ""
+        if data.get("age_sec") not in (None, ""):
+            age_text = f" age={data.get('age_sec')}"
+        return (
+            f"[XIAONIU] BLOCKED_REPLY room={room} user={sender} "
+            f"reason={data.get('reason', '')} "
+            f"trigger={data.get('trigger_type', 'none')} "
+            f"mode={data.get('reply_mode', '')} "
+            f"topic={data.get('topic', '-') or '-'} "
+            f"reply={data.get('response_preview', '')}"
+            f"{age_text}"
+        ).strip()
+
     if event == "sent":
         return (
             f"[XIAONIU] SENT room={room} user={sender_name}/{sender} "