[router] fix req handling order, improve serialization, remove retry (#8888)

sgl-router/scripts/run_benchmarks.py

@@ -121,6 +121,8 @@ class BenchmarkRunner:
                     results["serialization_time"] = self._extract_time(line)
                 elif "Deserialization (avg):" in line:
                     results["deserialization_time"] = self._extract_time(line)
+                elif "Bootstrap Injection (avg):" in line:
+                    results["bootstrap_injection_time"] = self._extract_time(line)
                 elif "Total Pipeline (avg):" in line:
                     results["total_time"] = self._extract_time(line)
 
@@ -143,6 +145,7 @@ class BenchmarkRunner:
         thresholds = {
             "serialization_time": 2000,  # 2μs max
             "deserialization_time": 2000,  # 2μs max
+            "bootstrap_injection_time": 5000,  # 5μs max
             "total_time": 10000,  # 10μs max
         }