feat：优化流式输出，增加临时多轮对话记录

app/config.py

@@ -50,7 +50,20 @@ class Settings:
         default_factory=lambda: int(os.getenv("MAX_ACTION_RETRIES", "2"))
     )
 
+    # 没有任何对话的空会话保留时长（秒），超时后视为不存在
+    conversation_empty_ttl: int = field(
+        default_factory=lambda: int(os.getenv("CONVERSATION_EMPTY_TTL", "7200"))
+    )
+    # 每次请求带入模型的最近对话轮数
+    conversation_history_turns: int = field(
+        default_factory=lambda: int(os.getenv("CONVERSATION_HISTORY_TURNS", "6"))
+    )
+
     def __post_init__(self):
+        if self.conversation_empty_ttl <= 0:
+            raise ValueError("CONVERSATION_EMPTY_TTL 必须是正整数")
+        if self.conversation_history_turns < 0:
+            raise ValueError("CONVERSATION_HISTORY_TURNS 必须是非负整数")
         if type(self.max_action_retries) is not int or self.max_action_retries < 0:
             raise ValueError("MAX_ACTION_RETRIES 必须是非负整数")
         if not 0 <= self.intent_min_confidence <= 1: