feat: add AI endpoint manager plugin with configuration and routing capabilities

- Introduced `ai_endpoint_mgr` plugin to manage multiple AI provider endpoints.
- Added configuration reference documentation for `config.toml`.
- Implemented endpoint loading, active endpoint switching, and model mutation.
- Included error handling for missing endpoints and configuration failures.
- Developed unit tests covering various scenarios including error paths and concurrency.

dstalk_web/src/main.cpp

@@ -43,6 +43,7 @@ class SseSession;
 // 插件服务 vtable 的全局指针，在启动时从主机查询获取。
 static const dstalk_ai_service_t*      g_ai      = nullptr;
 static const dstalk_session_service_t* g_session = nullptr;
+static const dstalk_ai_endpoint_mgr_t* g_endpoint_mgr = nullptr;  // I08: AI endpoint manager（可选）/ optional
 
 // ---- 运行时状态 / Runtime state ----
 // g_quit signals the main loop to exit (set by Ctrl+C).
@@ -208,8 +209,19 @@ static void run_chat_worker(
     };
 
     // 调用流式 AI 聊天 / Call streaming AI chat
-    dstalk_chat_result_t result = g_ai->chat_stream(
-        history, history_count, nullptr, token_cb, &cb_data);
+    // I08: 优先通过 endpoint_mgr 路由，fallback 到 g_ai / prefer endpoint_mgr, fallback to g_ai
+    dstalk_chat_result_t result = {};
+    const bool use_mgr = (g_endpoint_mgr && g_endpoint_mgr->count() > 0);
+    if (use_mgr) {
+        result = g_endpoint_mgr->chat_stream(
+            nullptr, history, history_count, nullptr, token_cb, &cb_data);
+    } else if (g_ai) {
+        result = g_ai->chat_stream(
+            history, history_count, nullptr, token_cb, &cb_data);
+    } else {
+        result.ok = 0;
+        result.error = dstalk_strdup("AI service unavailable");
+    }
 
     // 将 AI 回复加入会话 / Add AI reply to session
     if (result.ok) {
@@ -221,7 +233,11 @@ static void run_chat_worker(
     bool ok = result.ok;
     std::string content_copy = result.content ? result.content : "";
     std::string error_copy   = result.error  ? result.error  : "";
-    g_ai->free_result(&result);
+
+    // I08: 根据路由来源释放 result / free result based on routing source
+    if (use_mgr) g_endpoint_mgr->free_result(&result);
+    else if (g_ai) g_ai->free_result(&result);
+    else if (result.error) dstalk_free((void*)result.error);
 
     asio::post(ioc, [weak_sse, ok, content_copy, error_copy]() {
         if (auto sse = weak_sse.lock()) {
@@ -373,6 +389,23 @@ private:
         if (model) st["model"] = std::string(model);
         st["status"] = "running";
 
+        // I08/I09: endpoint manager 状态 / endpoint manager status
+        if (g_endpoint_mgr) {
+            st["endpoint_mgr_available"] = true;
+            st["endpoint_count"] = g_endpoint_mgr->count();
+            const char* active = g_endpoint_mgr->get_active();
+            if (active) st["active_endpoint"] = std::string(active);
+            char* list_json = g_endpoint_mgr->list_json();
+            if (list_json) {
+                boost::system::error_code ec;
+                auto jv = boost::json::parse(list_json, ec);
+                if (!ec) st["endpoints"] = std::move(jv);
+                dstalk_free(list_json);
+            }
+        } else {
+            st["endpoint_mgr_available"] = false;
+        }
+
         auto self = shared_from_this();
         http::response<http::string_body> res{http::status::ok, request_.version()};
         res.set("Access-Control-Allow-Origin", "*");
@@ -510,6 +543,9 @@ int main(int argc, char* argv[])
     if (!ai_provider) ai_provider = "ai_openai";
     g_ai      = static_cast<const dstalk_ai_service_t*>(dstalk_service_query(ai_provider, 1));
     g_session = static_cast<const dstalk_session_service_t*>(dstalk_service_query("session", 1));
+    // I08: 查询 AI endpoint manager（可选服务）/ query AI endpoint manager (optional service)
+    g_endpoint_mgr = static_cast<const dstalk_ai_endpoint_mgr_t*>(
+                         dstalk_service_query("ai_endpoint_mgr", 1));
 
     if (!g_ai) {
         std::fprintf(stderr, "[dstalk_web] AI service not found (check plugins directory)\n");