[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/cub/util_policy_wrapper_t.cuh

@@ -0,0 +1,31 @@
+// SPDX-FileCopyrightText: Copyright (c) 2011, Duane Merrill. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2011-2020, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+#pragma once
+
+#include <cub/config.cuh>
+
+CUB_NAMESPACE_BEGIN
+
+#ifndef _CCCL_DOXYGEN_INVOKED // Do not document
+
+namespace detail
+{
+/**
+ * @brief Helper class template that allows overwriting the `BLOCK_THREAD` and `ITEMS_PER_THREAD`
+ * configurations of a given policy.
+ */
+// TODO(bgruber): this should be called something like "override_policy"
+template <typename PolicyT, int BLOCK_THREADS_, int ITEMS_PER_THREAD_ = PolicyT::ITEMS_PER_THREAD>
+struct policy_wrapper_t : PolicyT
+{
+  static constexpr int ITEMS_PER_THREAD = ITEMS_PER_THREAD_;
+  static constexpr int BLOCK_THREADS    = BLOCK_THREADS_;
+  static constexpr int ITEMS_PER_TILE   = BLOCK_THREADS * ITEMS_PER_THREAD;
+};
+} // namespace detail
+
+#endif // _CCCL_DOXYGEN_INVOKED
+
+CUB_NAMESPACE_END