[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/benchmarks/bench/equal/basic.cu

@@ -0,0 +1,38 @@
+// SPDX-FileCopyrightText: Copyright (c) 2024, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include <thrust/device_vector.h>
+#include <thrust/equal.h>
+#include <thrust/execution_policy.h>
+
+#include "nvbench_helper.cuh"
+
+template <typename T>
+static void benchmark(nvbench::state& state, nvbench::type_list<T>)
+{
+  const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
+  thrust::device_vector<T> a(elements, T{1});
+  thrust::device_vector<T> b(elements, T{1});
+
+  const auto common_prefix = state.get_float64("CommonPrefixRatio");
+  const auto same_elements =
+    std::min(static_cast<std::size_t>(static_cast<double>(elements) * common_prefix), elements);
+  caching_allocator_t alloc;
+  thrust::fill(policy(alloc), b.begin() + same_elements, b.end(), T{2});
+
+  state.add_element_count(elements);
+  state.add_global_memory_reads<T>(2 * std::max(same_elements, std::size_t(1))); // using `same_elements` instead
+                                                                                 // of `elements` corresponds to the
+                                                                                 // actual elements read in an early
+                                                                                 // exit
+  state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::no_batch | nvbench::exec_tag::sync,
+             [&](nvbench::launch& launch) {
+               do_not_optimize(thrust::equal(policy(alloc, launch), a.begin(), a.end(), b.begin()));
+             });
+}
+
+NVBENCH_BENCH_TYPES(benchmark, NVBENCH_TYPE_AXES(integral_types))
+  .set_name("base")
+  .set_type_axes_names({"T{ct}"})
+  .add_int64_power_of_two_axis("Elements", nvbench::range(16, 28, 4))
+  .add_float64_axis("CommonPrefixRatio", std::vector{1.0, 0.5, 0.0});