feat(llm): 实现 Phase 0 剩余四个模块

实现 ProviderRegistry、HookExecutor、StreamEvents 和 Auto-compaction 模块，并集成到 LlmCycle 中

src/llm/provider.rs

@@ -1,7 +1,12 @@
 pub mod openai;
+pub mod registry;
+
+use std::pin::Pin;
+
+use tokio_stream::Stream;
 
 use crate::llm::error::LlmError;
-use crate::llm::types::{ChatRequest, ChatResponse};
+use crate::llm::types::{ChatRequest, ChatResponse, OpenaiChatChunk};
 use async_trait::async_trait;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -57,4 +62,19 @@ pub fn create_provider(
 pub trait LlmProvider: Send + Sync {
     /// 发送聊天请求并返回完整响应。
     async fn chat(&self, request: ChatRequest) -> Result<ChatResponse, LlmError>;
+
+    /// 流式聊天请求 —— 返回原始 SSE chunk 流。
+    ///
+    /// 默认实现回退到非流式调用（包装为单元素流）。
+    async fn chat_stream(
+        &self,
+        request: ChatRequest,
+    ) -> Result<
+        Pin<Box<dyn Stream<Item = Result<OpenaiChatChunk, LlmError>> + Send>>,
+        LlmError,
+    > {
+        let response = self.chat(request).await?;
+        let chunk = OpenaiChatChunk::from(response);
+        Ok(Box::pin(tokio_stream::once(Ok(chunk))))
+    }
 }