[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/cudax/examples/stf/01-axpy-cuda_kernel_chain.cu

@@ -0,0 +1,80 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDASTF in CUDA C++ Core Libraries,
+// under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+/**
+ * @file
+ *
+ * @brief Example of task implementing a chain of CUDA kernels
+ *
+ */
+
+#include <cuda/experimental/stf.cuh>
+
+using namespace cuda::experimental::stf;
+
+__global__ void axpy(double a, slice<const double> x, slice<double> y)
+{
+  int tid      = blockIdx.x * blockDim.x + threadIdx.x;
+  int nthreads = gridDim.x * blockDim.x;
+
+  for (int i = tid; i < x.size(); i += nthreads)
+  {
+    y(i) += a * x(i);
+  }
+}
+
+double X0(int i)
+{
+  return sin((double) i);
+}
+
+double Y0(int i)
+{
+  return cos((double) i);
+}
+
+int main()
+{
+  context ctx    = graph_ctx();
+  const size_t N = 16;
+  double X[N], Y[N];
+
+  for (size_t i = 0; i < N; i++)
+  {
+    X[i] = X0(i);
+    Y[i] = Y0(i);
+  }
+
+  double alpha = 3.14;
+  double beta  = 4.5;
+  double gamma = -4.1;
+
+  auto lX = ctx.logical_data(X);
+  auto lY = ctx.logical_data(Y);
+
+  /* Compute Y = Y + alpha X, Y = Y + beta X and then  Y = Y + gamma X */
+  ctx.cuda_kernel_chain(lX.read(), lY.rw())->*[&](auto dX, auto dY) {
+    // clang-format off
+        return std::vector<cuda_kernel_desc> {
+            { axpy, 16, 128, 0, alpha, dX, dY },
+            { axpy, 16, 128, 0, beta, dX, dY },
+            { axpy, 16, 128, 0, gamma, dX, dY }
+        };
+    // clang-format on
+  };
+
+  ctx.finalize();
+
+  for (size_t i = 0; i < N; i++)
+  {
+    assert(fabs(Y[i] - (Y0(i) + (alpha + beta + gamma) * X0(i))) < 0.0001);
+    assert(fabs(X[i] - X0(i)) < 0.0001);
+  }
+}