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

sgl-router/src/policies/cache_aware.rs

@@ -230,6 +230,10 @@ impl LoadBalancingPolicy for CacheAwarePolicy {
         "cache_aware"
     }
 
+    fn needs_request_text(&self) -> bool {
+        true // Cache-aware policy needs request text for cache affinity
+    }
+
     fn on_request_complete(&self, worker_url: &str, success: bool) {
         // Could track success rates per worker for more intelligent routing
         if !success {

sgl-router/src/policies/mod.rs

@@ -59,6 +59,11 @@ pub trait LoadBalancingPolicy: Send + Sync + Debug {
     /// Get policy name for metrics and debugging
     fn name(&self) -> &'static str;
 
+    /// Check if this policy needs request text for routing decisions
+    fn needs_request_text(&self) -> bool {
+        false // Default: most policies don't need request text
+    }
+
     /// Update worker load information
     ///
     /// This is called periodically with current load information for load-aware policies.