[CCCL] Add missing CCCL components: c2h, nvbench_helper, cmake, cudax, AGENTS.md

Added 863 files from NVIDIA/cccl sparse checkout:
- c2h/ (27 files): Catch2 test helpers — generators, validators, runner
- nvbench_helper/ (10 files): Benchmark harness utilities
- cmake/ (29 files): CMake presets and build helpers
- cudax/ (794 files): Experimental CUDA extensions
- AGENTS.md: NVIDIA's official AI agent instructions for CCCL
- CMakePresets.json: Standardized build configurations
- cccl-version.json: Version tracking

Also added CCCL_ASSET_MAP.md mapping all 4295 CCCL files to
competition value and PRD items.

cccl_upstream now covers 100% of competition-critical assets:
- 27 tuning headers (SM80/90/100 benchmark data)
- 32 dispatch headers (algorithm implementations)
- 60 Thrust examples (correctness verification)
- 217 CUB Catch2 tests (regression matrix)
- 153 CUB benchmarks (parameter space search)
- 18 CUB examples (API verification)
- 27 test helpers + benchmark harness
- 794 cudax experimental extensions

cccl_upstream/cudax/test/stf/stress/kernel_chain.cu

@@ -0,0 +1,72 @@
+//===----------------------------------------------------------------------===//
+//
+// 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.
+//
+//===----------------------------------------------------------------------===//
+
+#include <cuda/experimental/stf.cuh>
+
+using namespace cuda::experimental::stf;
+
+__global__ void swap_kernel(slice<double> dst, slice<double> src)
+{
+  size_t tid      = threadIdx.x + blockIdx.x * blockDim.x;
+  size_t nthreads = blockDim.x * gridDim.x;
+  size_t n        = dst.size();
+
+  for (size_t i = tid; i < n; i += nthreads)
+  {
+    double tmp = dst(i);
+    dst(i)     = src(i);
+    src(i)     = tmp;
+  }
+}
+
+int main(int argc, char** argv)
+{
+  stream_ctx ctx;
+  const size_t N = 16;
+  double X[N], Y[N];
+
+  for (size_t i = 0; i < N; i++)
+  {
+    X[i] = 1.0;
+    Y[i] = 2.0;
+  }
+
+  auto lX = ctx.logical_data(X);
+  auto lY = ctx.logical_data(Y);
+
+#ifdef NDEBUG
+  size_t iter_cnt = 10000000;
+#else
+  size_t iter_cnt = 10000;
+  fprintf(stderr, "Warning: Running with small problem size in debug mode, should use DEBUG=0.\n");
+#endif
+
+  if (argc > 1)
+  {
+    iter_cnt = atol(argv[1]);
+  }
+
+  std::chrono::steady_clock::time_point start, stop;
+  start = std::chrono::steady_clock::now();
+  for (size_t iter = 0; iter < iter_cnt; iter++)
+  {
+    ctx.task(lX.rw(), lY.rw())->*[&](cudaStream_t s, auto dX, auto dY) {
+      swap_kernel<<<4, 16, 0, s>>>(dY, dX);
+    };
+    ctx.task(lY.rw(), lX.rw())->*[&](cudaStream_t s, auto dY, auto dX) {
+      swap_kernel<<<4, 16, 0, s>>>(dX, dY);
+    };
+  }
+  stop = std::chrono::steady_clock::now();
+  ctx.finalize();
+
+  std::chrono::duration<double> duration = stop - start;
+  fprintf(stderr, "Elapsed: %.2lf us per task pair\n", duration.count() * 1000000.0 / (iter_cnt));
+}