fix: adapt xllm norm.cu for corex CUB (CUDA 10.2)

Key change: replace CCCL 3.6 types with corex CUB equivalents
  - cuda::std::plus<> → cub::Sum
  - cuda::maximum<>  → cub::Max
  - Remove #include <cuda/std/functional>

Test results from real machine (3/4 passed):
  ✓ __shfl_down_sync works on ivcore10
  ✓ manual SMEM+shuffle block reduce works
  ✓ corex CUB cub::BlockReduce<float,256> compiles and runs correctly (32640)
  ✗ CCCL 3.6 variadic function issue — corex clang rejects device variadic

Confirmed: use /usr/local/corex/include/cub/ for all kernel code
           cccl_upstream is reference only, NOT compilable on corex

Build script: bash qwen3_6_scripts/build_xllm_kernels.sh

ex_engine/xllm_kernels/cuda/norm.cu

@@ -28,9 +28,9 @@ limitations under the License.
 // https://github.com/vllm-project/vllm/blob/main/csrc/layernorm_kernels.cu
 
 #if CUB_VERSION >= 200800
-#include <cuda/std/functional>
-using CubAddOp = ::cuda::std::plus<>;
-using CubMaxOp = ::cuda::maximum<>;
+// corex CUB (CUDA 10.2) — no cuda::std::functional
+using CubAddOp = cub::Sum;
+using CubMaxOp = cub::Max;
 #else   // if CUB_VERSION < 200800
 using CubAddOp = cub::Sum;
 using CubMaxOp = cub::Max;