[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/thrust/thrust/detail/count.inl

@@ -0,0 +1,88 @@
+// SPDX-FileCopyrightText: Copyright (c) 2008-2013, NVIDIA Corporation. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+#pragma once
+
+#include <thrust/detail/config.h>
+
+#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
+#  pragma GCC system_header
+#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
+#  pragma clang system_header
+#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
+#  pragma system_header
+#endif // no system header
+#include <thrust/count.h>
+#include <thrust/iterator/iterator_traits.h>
+#include <thrust/system/detail/generic/select_system.h>
+
+// Include all active backend system implementations (generic, sequential, host and device)
+#include <thrust/system/detail/generic/count.h>
+#include <thrust/system/detail/sequential/count.h>
+#include __THRUST_HOST_SYSTEM_ALGORITH_DETAIL_HEADER_INCLUDE(count.h)
+#include __THRUST_DEVICE_SYSTEM_ALGORITH_DETAIL_HEADER_INCLUDE(count.h)
+
+// Some build systems need a hint to know which files we could include
+#if 0
+#  include <thrust/system/cpp/detail/count.h>
+#  include <thrust/system/cuda/detail/count.h>
+#  include <thrust/system/omp/detail/count.h>
+#  include <thrust/system/tbb/detail/count.h>
+#endif
+
+THRUST_NAMESPACE_BEGIN
+
+_CCCL_EXEC_CHECK_DISABLE
+template <typename DerivedPolicy, typename InputIterator, typename EqualityComparable>
+_CCCL_HOST_DEVICE thrust::detail::it_difference_t<InputIterator>
+count(const thrust::detail::execution_policy_base<DerivedPolicy>& exec,
+      InputIterator first,
+      InputIterator last,
+      const EqualityComparable& value)
+{
+  _CCCL_NVTX_RANGE_SCOPE("thrust::count");
+  using thrust::system::detail::generic::count;
+  return count(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, value);
+} // end count()
+
+_CCCL_EXEC_CHECK_DISABLE
+template <typename DerivedPolicy, typename InputIterator, typename Predicate>
+_CCCL_HOST_DEVICE thrust::detail::it_difference_t<InputIterator>
+count_if(const thrust::detail::execution_policy_base<DerivedPolicy>& exec,
+         InputIterator first,
+         InputIterator last,
+         Predicate pred)
+{
+  _CCCL_NVTX_RANGE_SCOPE("thrust::count_if");
+  using thrust::system::detail::generic::count_if;
+  return count_if(thrust::detail::derived_cast(thrust::detail::strip_const(exec)), first, last, pred);
+} // end count_if()
+
+template <typename InputIterator, typename EqualityComparable>
+thrust::detail::it_difference_t<InputIterator>
+count(InputIterator first, InputIterator last, const EqualityComparable& value)
+{
+  _CCCL_NVTX_RANGE_SCOPE("thrust::count");
+  using thrust::system::detail::generic::select_system;
+
+  using System = typename thrust::iterator_system<InputIterator>::type;
+
+  System system;
+
+  return thrust::count(select_system(system), first, last, value);
+} // end count()
+
+template <typename InputIterator, typename Predicate>
+thrust::detail::it_difference_t<InputIterator> count_if(InputIterator first, InputIterator last, Predicate pred)
+{
+  _CCCL_NVTX_RANGE_SCOPE("thrust::count_if");
+  using thrust::system::detail::generic::select_system;
+
+  using System = typename thrust::iterator_system<InputIterator>::type;
+
+  System system;
+
+  return thrust::count_if(select_system(system), first, last, pred);
+} // end count_if()
+
+THRUST_NAMESPACE_END