[router] create worker removal step and clean up worker manager (#11921)

sgl-router/src/routers/grpc/responses/conversions.rs

@@ -149,7 +149,11 @@ pub fn responses_to_chat(req: &ResponsesRequest) -> Result<ChatCompletionRequest
 
     Ok(ChatCompletionRequest {
         messages,
-        model: req.model.clone().unwrap_or_else(|| "default".to_string()),
+        model: if req.model.is_empty() {
+            "default".to_string()
+        } else {
+            req.model.clone()
+        },
         temperature: req.temperature,
         max_completion_tokens: req.max_output_tokens,
         stream: is_streaming,
@@ -311,7 +315,7 @@ mod tests {
         let req = ResponsesRequest {
             input: ResponseInput::Text("Hello, world!".to_string()),
             instructions: Some("You are a helpful assistant.".to_string()),
-            model: Some("gpt-4".to_string()),
+            model: "gpt-4".to_string(),
             temperature: Some(0.7),
             ..Default::default()
         };

sgl-router/src/routers/grpc/responses/handlers.rs

@@ -324,10 +324,11 @@ async fn route_responses_background(
         incomplete_details: None,
         instructions: request.instructions.clone(),
         max_output_tokens: request.max_output_tokens,
-        model: request
-            .model
-            .clone()
-            .unwrap_or_else(|| "default".to_string()),
+        model: if request.model.is_empty() {
+            "default".to_string()
+        } else {
+            request.model.clone()
+        },
         output: Vec::new(),
         parallel_tool_calls: request.parallel_tool_calls.unwrap_or(true),
         previous_response_id: request.previous_response_id.clone(),
@@ -622,10 +623,11 @@ async fn process_and_transform_sse_stream(
 
     // Create event emitter for OpenAI-compatible streaming
     let response_id = format!("resp_{}", Uuid::new_v4());
-    let model = original_request
-        .model
-        .clone()
-        .unwrap_or_else(|| "default".to_string());
+    let model = if original_request.model.is_empty() {
+        "default".to_string()
+    } else {
+        original_request.model.clone()
+    };
     let created_at = chrono::Utc::now().timestamp() as u64;
     let mut event_emitter = ResponseStreamEventEmitter::new(response_id, model, created_at);
 

sgl-router/src/routers/grpc/responses/tool_loop.rs

@@ -608,10 +608,11 @@ async fn execute_tool_loop_streaming_internal(
 
     // Create response event emitter
     let response_id = format!("resp_{}", Uuid::new_v4());
-    let model = current_request
-        .model
-        .clone()
-        .unwrap_or_else(|| "default".to_string());
+    let model = if current_request.model.is_empty() {
+        "default".to_string()
+    } else {
+        current_request.model.clone()
+    };
     let created_at = SystemTime::now()
         .duration_since(UNIX_EPOCH)
         .unwrap()