feat(prompt): 添加提示词工程模块并扩展LLM周期接口

新增 `prompt` 模块，包含模板引擎、组合器和错误类型，同时在`LlmCycle`中增加直接操作消息历史的方法和`submit_messages`接口

src/llm/cycle.rs

@@ -113,6 +113,94 @@ impl LlmCycle {
         self.usage.reset();
     }
 
+    /// 直接设置消息历史（覆盖已有消息），支持 Builder 链式调用。
+    pub fn with_messages(mut self, messages: Vec<OpenaiChatMessage>) -> Self {
+        self.messages = messages;
+        self
+    }
+
+    /// 追加消息到历史尾部。
+    pub fn extend_messages(&mut self, messages: Vec<OpenaiChatMessage>) {
+        self.messages.extend(messages);
+    }
+
+    /// 使用预构建消息提交（跳过自动 push user prompt）。
+    ///
+    /// 与 `submit()` 不同，不自动添加 `user_text(prompt)`，也不自动插入 system prompt。
+    /// 调用方完全控制消息序列内容。
+    pub async fn submit_messages(
+        &mut self,
+        messages: Vec<OpenaiChatMessage>,
+        tools: Vec<ToolDefinition>,
+    ) -> Result<ChatResponse, LlmError> {
+        let openai_tools: Option<Vec<OpenaiTool>> = if tools.is_empty() {
+            None
+        } else {
+            Some(
+                tools
+                    .iter()
+                    .map(|t| OpenaiTool::Function {
+                        function: t.clone(),
+                    })
+                    .collect(),
+            )
+        };
+
+        let request = ChatRequest {
+            model: self.config.model.clone(),
+            messages,
+            max_tokens: self.config.max_tokens,
+            temperature: self.config.temperature,
+            tools: openai_tools,
+            tool_choice: Some(ToolChoice::Auto),
+            ..Default::default()
+        };
+
+        if let Some(ref executor) = self.hook_executor {
+            let ctx = HookContext::new(crate::llm::hooks::HookEvent::PreRequest)
+                .with_request(&request);
+            let results = executor
+                .execute(crate::llm::hooks::HookEvent::PreRequest, &ctx)
+                .await;
+            if results.iter().any(|r| r.should_block) {
+                let reason = results
+                    .iter()
+                    .find(|r| r.should_block)
+                    .and_then(|r| r.reason.clone())
+                    .unwrap_or_else(|| "Blocked by pre-request hook".to_string());
+                return Err(LlmError::Other(reason));
+            }
+        }
+
+        match self.provider.chat(request).await {
+            Ok(response) => {
+                if let Some(ref executor) = self.hook_executor {
+                    let post_request = ChatRequest {
+                        model: self.config.model.clone(),
+                        messages: vec![],
+                        ..Default::default()
+                    };
+                    let ctx = HookContext::new(crate::llm::hooks::HookEvent::PostRequest)
+                        .with_request(&post_request);
+                    executor
+                        .execute(crate::llm::hooks::HookEvent::PostRequest, &ctx)
+                        .await;
+                }
+                self.usage.add(&response.usage);
+                Ok(response)
+            }
+            Err(e) => {
+                if let Some(ref executor) = self.hook_executor {
+                    let ctx = HookContext::new(crate::llm::hooks::HookEvent::OnError).with_error(&e);
+                    executor
+                        .execute(crate::llm::hooks::HookEvent::OnError, &ctx)
+                        .await;
+                }
+                Err(e)
+            }
+        }
+    }
+
     /// 提交用户消息并获取 LLM 响应。
     pub async fn submit(
         &mut self,