[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/test/stf/local_stf/stackable.cu

@@ -0,0 +1,88 @@
+//===----------------------------------------------------------------------===//
+//
+// 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 Experiment with local context nesting
+ *
+ */
+
+#include <cuda/experimental/stf.cuh>
+
+using namespace cuda::experimental::stf;
+
+int main()
+{
+  stackable_ctx sctx;
+
+  int array[1024];
+  for (size_t i = 0; i < 1024; i++)
+  {
+    array[i] = 1 + i * i;
+  }
+
+  auto lC = sctx.logical_data(array);
+
+  auto lA = sctx.logical_data(lC.shape());
+  lA.set_symbol("A");
+
+  auto lA2 = sctx.logical_data(shape_of<slice<int>>(1024));
+  lA2.set_symbol("A2");
+
+  sctx.parallel_for(lA.shape(), lA.write())->*[] __device__(size_t i, auto a) {
+    a(i) = 42 + 2 * i;
+  };
+
+  /* Create nested graph */
+  {
+    stackable_ctx::graph_scope_guard scope{sctx};
+
+    auto lB = sctx.logical_data(shape_of<slice<int>>(512));
+    lB.set_symbol("B");
+
+    sctx.parallel_for(lB.shape(), lB.write())->*[] __device__(size_t i, auto b) {
+      b(i) = 17 - 3 * i;
+    };
+
+    sctx.parallel_for(lA2.shape(), lA2.write())->*[] __device__(size_t i, auto a2) {
+      a2(i) = 5 * i + 4;
+    };
+
+    sctx.parallel_for(lB.shape(), lA.read(), lB.rw())->*[] __device__(size_t i, auto a, auto b) {
+      b(i) += a(i);
+    };
+
+    sctx.parallel_for(lB.shape(), lB.read(), lC.rw())->*[] __device__(size_t i, auto b, auto c) {
+      c(i) += b(i);
+    };
+  }
+
+  sctx.host_launch(lA2.read())->*[](auto a2) {
+    for (size_t i = 0; i < a2.size(); i++)
+    {
+      EXPECT(a2(i) == 5 * i + 4);
+    }
+  };
+
+  // Do the same check in another graph
+  {
+    stackable_ctx::graph_scope_guard scope{sctx};
+    lA2.push(access_mode::read);
+    sctx.host_launch(lA2.read())->*[](auto a2) {
+      for (size_t i = 0; i < a2.size(); i++)
+      {
+        EXPECT(a2(i) == 5 * i + 4);
+      }
+    };
+  }
+
+  sctx.finalize();
+}