feat(core): 完成 Phase 11 测试与检索补强

- 新增 VectorRetriever trait 与 InMemoryVectorRetriever 引用实现
- 补充 Provider roundtrip wiremock 测试与 MemoryStore 并发测试共 23 个
- 修复 openai 429 retry-after header 解析（与 anthropic 对齐）

src/llm/provider/anthropic.rs

@@ -901,7 +901,7 @@ mod tests {
     use super::*;
     use crate::llm::types::request_v2::MessageRequest;
     use serde_json::json;
-    use wiremock::matchers::{method, path};
+    use wiremock::matchers::{header, method, path};
     use wiremock::{Mock, MockServer, ResponseTemplate};
 
     fn make_provider(base_url: String) -> AnthropicProvider {
@@ -1077,4 +1077,134 @@ event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n";
         assert_eq!(body.max_tokens, DEFAULT_MAX_TOKENS);
         assert_eq!(body.model, "claude-sonnet-4-20250514");
     }
+
+    // ===== Phase 11 Step 11.2 wiremock roundtrip 测试 =====
+
+    #[tokio::test]
+    async fn anthropic_401_structured_error() {
+        let server = MockServer::start().await;
+        Mock::given(method("POST"))
+            .and(path("/v1/messages"))
+            .respond_with(ResponseTemplate::new(401).set_body_json(json!({
+                "type": "error",
+                "error": {
+                    "type": "authentication_error",
+                    "message": "Invalid API key provided: sk-ant-test"
+                }
+            })))
+            .mount(&server)
+            .await;
+
+        let provider = make_provider(server.uri());
+        let err = provider
+            .chat_blocking(MessageRequest {
+                model: "claude-sonnet-4-20250514".into(),
+                messages: vec![Message::user_text("Hi")],
+                ..Default::default()
+            })
+            .await
+            .unwrap_err();
+        match err {
+            LlmError::Authentication(msg) => assert!(msg.contains("Invalid API key")),
+            other => panic!("expected Authentication, got {other:?}"),
+        }
+    }
+
+    #[tokio::test]
+    async fn anthropic_tool_use_response() {
+        let server = MockServer::start().await;
+        Mock::given(method("POST"))
+            .and(path("/v1/messages"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+                "id": "msg_tool",
+                "type": "message",
+                "model": "claude-sonnet-4-20250514",
+                "content": [
+                    {"type": "text", "text": "Let me check."},
+                    {"type": "tool_use", "id": "toolu_abc", "name": "lookup", "input": {"q": "rust"}}
+                ],
+                "stop_reason": "tool_use",
+                "usage": {"input_tokens": 8, "output_tokens": 12}
+            })))
+            .mount(&server)
+            .await;
+
+        let provider = make_provider(server.uri());
+        let response = provider
+            .chat_blocking(MessageRequest {
+                model: "claude-sonnet-4-20250514".into(),
+                messages: vec![Message::user_text("Look up rust")],
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+        assert_eq!(response.stop_reason, StopReason::ToolUse);
+        let tool_use = match &response.message {
+            Message::Assistant { content } => content.iter().find_map(|b| match b {
+                ContentBlock::ToolUse { id, name, .. } => Some((id.clone(), name.clone())),
+                _ => None,
+            }),
+            _ => None,
+        };
+        assert_eq!(tool_use, Some(("toolu_abc".into(), "lookup".into())));
+    }
+
+    #[tokio::test]
+    async fn anthropic_version_header() {
+        let server = MockServer::start().await;
+        Mock::given(method("POST"))
+            .and(path("/v1/messages"))
+            .and(header("anthropic-version", "2023-06-01"))
+            .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+                "id": "msg_v",
+                "type": "message",
+                "model": "claude-sonnet-4-20250514",
+                "content": [{"type": "text", "text": "OK"}],
+                "stop_reason": "end_turn",
+                "usage": {"input_tokens": 1, "output_tokens": 1}
+            })))
+            .mount(&server)
+            .await;
+
+        let provider = make_provider(server.uri());
+        let response = provider
+            .chat_blocking(MessageRequest {
+                model: "claude-sonnet-4-20250514".into(),
+                messages: vec![Message::user_text("Hi")],
+                ..Default::default()
+            })
+            .await
+            .unwrap();
+        assert_eq!(response.text(), "OK");
+    }
+
+    #[tokio::test]
+    async fn anthropic_529_overloaded_structured() {
+        let server = MockServer::start().await;
+        Mock::given(method("POST"))
+            .and(path("/v1/messages"))
+            .respond_with(ResponseTemplate::new(529).set_body_json(json!({
+                "type": "error",
+                "error": {
+                    "type": "overloaded_error",
+                    "message": "Overloaded: Anthropic API is temporarily overloaded"
+                }
+            })))
+            .mount(&server)
+            .await;
+
+        let provider = make_provider(server.uri());
+        let err = provider
+            .chat_blocking(MessageRequest {
+                model: "claude-sonnet-4-20250514".into(),
+                messages: vec![Message::user_text("Hi")],
+                ..Default::default()
+            })
+            .await
+            .unwrap_err();
+        match err {
+            LlmError::RateLimit { retry_after } => assert!(retry_after.is_none()),
+            other => panic!("expected RateLimit, got {other:?}"),
+        }
+    }
 }