[smg] cleanup router RAII guards (#16560)

sgl-model-gateway/src/routers/http/pd_router.rs

@@ -835,7 +835,7 @@ impl PDRouter {
         prefill: Arc<dyn Worker>,
         decode: Arc<dyn Worker>,
     ) -> Response {
-        use crate::core::attach_guards_to_response;
+        use crate::core::AttachedBody;
 
         let (tx, rx) = tokio::sync::mpsc::unbounded_channel();
 
@@ -885,7 +885,7 @@ impl PDRouter {
         // Attach load guards to response body for proper RAII lifecycle
         // Guards are dropped when response body is consumed or client disconnects
         let guards = vec![WorkerLoadGuard::new(prefill), WorkerLoadGuard::new(decode)];
-        attach_guards_to_response(guards, response)
+        AttachedBody::wrap_response(response, guards)
     }
 
     // Helper to process non-streaming decode response with logprob merging

sgl-model-gateway/src/routers/http/router.rs

@@ -16,7 +16,7 @@ use crate::{
     app_context::AppContext,
     config::types::RetryConfig,
     core::{
-        is_retryable_status, ConnectionMode, RetryExecutor, Worker, WorkerLoadGuard,
+        is_retryable_status, AttachedBody, ConnectionMode, RetryExecutor, Worker, WorkerLoadGuard,
         WorkerRegistry, WorkerType, UNKNOWN_MODEL_ID,
     },
     observability::{
@@ -629,7 +629,7 @@ impl Router {
             // Attach load guard to response body for proper RAII lifecycle
             // Guard is dropped when response body is consumed or client disconnects
             if let Some(guard) = load_guard {
-                response = guard.attach_to_response(response);
+                response = AttachedBody::wrap_response(response, guard);
             }
             response
         }