[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/include/cuda/experimental/stf.cuh

@@ -0,0 +1,42 @@
+//===----------------------------------------------------------------------===//
+//
+// 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) 2022-2025 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+/** @file
+ *
+ * @brief Main include file for the CUDASTF library.
+ *
+ * STF pulls `cuda::experimental::places` into `cuda::experimental::stf` via
+ * `__stf/internal/stf_places_into_stf_core.cuh` (after `places.cuh` in STF-only
+ * headers), `stf_places_extended_exports.cuh` (after `stf::hash`; avoids including
+ * `place_partition.cuh`), and `stf_places_partition_into_stf.cuh` where partition
+ * APIs are needed. Places headers do not include STF for that bridge.
+ */
+
+#pragma once
+
+#include <cuda/experimental/__stf/allocators/adapters.cuh>
+#include <cuda/experimental/__stf/allocators/buddy_allocator.cuh>
+#include <cuda/experimental/__stf/allocators/cached_allocator.cuh>
+#include <cuda/experimental/__stf/allocators/pooled_allocator.cuh>
+#include <cuda/experimental/__stf/allocators/uncached_allocator.cuh>
+#include <cuda/experimental/__stf/graph/graph_ctx.cuh>
+// #include <cuda/experimental/__stf/internal/algorithm.cuh>
+#include <cuda/experimental/__places/exec/cuda_context.cuh>
+#include <cuda/experimental/__places/exec/cuda_stream.cuh>
+#include <cuda/experimental/__places/exec/green_context.cuh>
+#include <cuda/experimental/__stf/internal/context.cuh>
+#include <cuda/experimental/__stf/internal/inner_shape.cuh>
+#include <cuda/experimental/__stf/internal/reducer.cuh>
+#include <cuda/experimental/__stf/internal/scalar_interface.cuh>
+#include <cuda/experimental/__stf/internal/task_dep.cuh>
+#include <cuda/experimental/__stf/internal/void_interface.cuh>
+#include <cuda/experimental/__stf/stackable/stackable_ctx.cuh>
+#include <cuda/experimental/__stf/stream/stream_ctx.cuh>
+#include <cuda/experimental/__stf/utility/run_once.cuh>