feat: support flashinfer kernel autotune (#12306)

Co-authored-by: Qiaolin Yu <liin1211@outlook.com>
Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>

python/sglang/srt/server_args.py

@@ -378,6 +378,7 @@ class ServerArgs:
     mm_attention_backend: Optional[str] = None
     nsa_prefill_backend: str = "flashmla_sparse"
     nsa_decode_backend: str = "fa3"
+    enable_flashinfer_autotune: bool = False
 
     # Speculative decoding
     speculative_algorithm: Optional[str] = None
@@ -2882,6 +2883,12 @@ class ServerArgs:
             type=str,
             choices=NSA_CHOICES,
         )
+        parser.add_argument(
+            "--enable-flashinfer-autotune",
+            default=ServerArgs.enable_flashinfer_autotune,
+            action="store_true",
+            help="Enable FlashInfer autotuning for optimal kernel selection.",
+        )
 
         # Speculative decoding
         parser.add_argument(