Fix FP4 MoE accuracy from missing routed_scaling_factor (#8333)

python/sglang/srt/server_args.py

@@ -433,10 +433,6 @@ class ServerArgs:
                 self.quantization == "modelopt_fp4"
             ), "modelopt_fp4 quantization is required for Flashinfer MOE"
             os.environ["TRTLLM_ENABLE_PDL"] = "1"
-            self.disable_shared_experts_fusion = True
-            logger.warning(
-                f"Flashinfer MoE is enabled. Shared expert fusion is disabled."
-            )
 
         # DeepEP MoE
         if self.enable_deepep_moe: