cuda13.0, implement changes to CCCL (#6751)

custom_ops/gpu_ops/quantization/common.cu

@@ -209,7 +209,8 @@ __global__ void dynamic_per_token_scaled_fp8_quant_kernel(
   using BlockReduce = cub::BlockReduce<float, 1024>;
   __shared__ typename BlockReduce::TempStorage reduceStorage;
   float const block_absmax_val_maybe =
-      BlockReduce(reduceStorage).Reduce(absmax_val, cub::Max{}, blockDim.x);
+      BlockReduce(reduceStorage)
+          .Reduce(absmax_val, fd_cub_compat::Max{}, blockDim.x);
   __shared__ float token_scale;
   if (tid == 0) {
     if (scale_ub > 0) {

custom_ops/gpu_ops/quantization/common.cuh

@@ -9,6 +9,8 @@
 #include <cub/cub.cuh>
 #include <cuda_runtime.h>
 
+#include "../cccl_compat.h"  // CCCL 3.0 compatibility
+
 namespace fastdeploy {
 
 // Vectorization containers