[EP] Add cuda kernel for moe_ep_pre_reorder (#6699)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>

sgl-kernel/include/sgl_kernel_ops.h

@@ -240,6 +240,17 @@ void prepare_moe_input(
     const int64_t n,
     const int64_t k);
 
+void ep_moe_pre_reorder(
+    torch::Tensor input,
+    torch::Tensor gateup_input,
+    torch::Tensor src2dst,
+    torch::Tensor topk_ids,
+    torch::Tensor a1_scales,
+    int64_t start_expert_id,
+    int64_t end_expert_id,
+    int64_t topk,
+    bool use_per_token_if_dynamic);
+
 /*
  * From csrc/speculative
  */