[INFRA] Import NVIDIA/CCCL upstream as optimization reference library

CCCL (CUDA C++ Core Libraries) provides:
- CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk)
- Thrust: high-level parallel algorithms (transform_reduce, sort, scan)
- libcudacxx: CUDA C++ standard library (atomics, barriers, memory)
- cudax: experimental features (memory resources, allocators)
- Tuning policies: per-SM hardware-specific algorithm parameters

Competition optimization vectors mapped to CCCL:
- Output TPS (83% weight): warp_reduce, block_reduce, device_topk
- Input TPS (14% weight): device_scan, block_load, prefetch
- Cache TPS (3% weight): prefix caching strategy patterns
- Memory (0.9 util): pooled/cached/buddy allocators

Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only)
License: Apache-2.0

cccl_upstream/cub/test/test_nvtx_standalone.cu

@@ -0,0 +1,21 @@
+// The purpose of this test is to verify that CUB can use NVTX without any additional dependencies. It is built as part
+// of the unit tests, but can also be built standalone:
+
+// Compile (from current directory):
+//   nvcc test_nvtx_standalone.cu -I../../cub -I../../thrust -I../../libcudacxx/include -o nvtx_standalone
+// Profile & view:
+//   (nsys profile -o nvtx_standalone.nsys-rep -f true ./nvtx_standalone || true) && nsys-ui nvtx_standalone.nsys-rep
+
+#include <cub/device/device_for.cuh>
+
+#include <cuda/iterator>
+#include <cuda/std/functional>
+
+int main()
+{
+  _CCCL_NVTX_RANGE_SCOPE("main");
+
+  cuda::counting_iterator<int> it{0};
+  cub::DeviceFor::ForEach(it, it + 16, ::cuda::std::negate<int>{});
+  cudaDeviceSynchronize();
+}