[CCCL] 瘦身 + 补全: 移除 cudax/python/libcudacxx-tests 冗余文件, 新增 c2h 测试助手 + cmake 构建系统 + 8 个 CUDA thrust examples

变更摘要:
- 删除: cudax/ (783 files, 7.2M) — 实验性组件，竞赛不需要
- 删除: python/ (226 files, 2.0M) — Python 绑定，竞赛不需要
- 删除: libcudacxx/{test,benchmarks,codegen,cmake,share} (4432 files, 31M)
  保留: libcudacxx/include/ (1463 headers, cuda::std 编译依赖)
- 新增: c2h/ (27 files) — CUB Catch2 测试辅助头文件，编译 243 个测试必需
- 新增: cmake/ (29 files) — CCCL 原生 CMake 构建系统
- 新增: thrust/examples/cuda/ (7 files) + cpp_integration/ (1 file)
  async_reduce, custom_temporary_allocation, explicit_cuda_stream,
  global_device_vector, range_view, unwrap_pointer, wrap_pointer, device

结果: cccl_upstream 从 74M→35M (瘦身 53%), 核心内容 100% 保留:
  27/27 tuning headers, 78 benchmarks, 243 tests,
  60 thrust examples, 18 CUB examples, 全部编译头文件

cccl_upstream/cudax/benchmarks/bench/cuco/hyperloglog.cu

@@ -1,133 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// Part of CUDA Experimental 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) 2026 NVIDIA CORPORATION & AFFILIATES.
-//
-//===----------------------------------------------------------------------===//
-
-#include <thrust/execution_policy.h>
-#include <thrust/sequence.h>
-
-#include <cuda/buffer>
-#include <cuda/memory_resource>
-#include <cuda/std/cmath>
-#include <cuda/std/cstddef>
-#include <cuda/stream>
-
-#include <cuda/experimental/__cuco/hyperloglog.cuh>
-
-#include "common/defaults.cuh"
-#include <nvbench/nvbench.cuh>
-
-namespace cudax = cuda::experimental;
-namespace bench = cudax::cuco::benchmark;
-
-namespace
-{
-template <typename Key>
-void add_relative_error_summary(
-  nvbench::state& state,
-  cudax::cuco::hyperloglog<Key>& estimator,
-  cuda::stream_ref stream,
-  Key* first,
-  cuda::std::size_t num_items)
-{
-  estimator.add(stream, first, first + num_items);
-  const auto estimated_cardinality = estimator.estimate(stream);
-  const auto relative_error =
-    cuda::std::abs(static_cast<double>(estimated_cardinality) / static_cast<double>(num_items) - 1.0);
-  estimator.clear(stream);
-
-  auto& summary = state.add_summary("RelativeError");
-  summary.set_string("hint", "RelErr");
-  summary.set_string("short_name", "RelativeError");
-  summary.set_string("description", "Relative approximation error.");
-  summary.set_float64("value", relative_error);
-}
-} // namespace
-
-/**
- * @brief A benchmark evaluating `cudax::cuco::hyperloglog` end-to-end performance.
- */
-template <typename Key>
-void hyperloglog_e2e(nvbench::state& state, nvbench::type_list<Key>)
-{
-  using estimator_type      = cudax::cuco::hyperloglog<Key>;
-  using sketch_size_kb_type = typename estimator_type::sketch_size_kb;
-
-  const auto num_items      = static_cast<cuda::std::size_t>(state.get_int64("NumInputs"));
-  const auto sketch_size_kb = sketch_size_kb_type{static_cast<double>(state.get_int64("SketchSizeKB"))};
-
-  const auto device = cuda::device_ref{0};
-  cuda::stream stream{device};
-  const cuda::device_memory_pool_ref mr = cuda::device_default_memory_pool(device);
-
-  auto items = cuda::make_device_buffer<Key>(stream, device, num_items, cuda::no_init);
-  thrust::sequence(thrust::cuda::par_nosync.on(stream.get()), items.begin(), items.end(), Key{0});
-
-  estimator_type estimator{stream, mr, sketch_size_kb};
-  stream.sync();
-
-  state.add_element_count(num_items);
-  state.add_global_memory_reads<Key>(num_items, "InputSize");
-
-  add_relative_error_summary(state, estimator, stream, items.data(), num_items);
-
-  state.exec(nvbench::exec_tag::sync | nvbench::exec_tag::timer, [&](nvbench::launch& launch, auto& timer) {
-    timer.start();
-    estimator.add_async({launch.get_stream()}, items.begin(), items.end());
-    [[maybe_unused]] const auto estimated_cardinality = estimator.estimate({launch.get_stream()});
-    timer.stop();
-
-    estimator.clear_async({launch.get_stream()});
-  });
-}
-
-/**
- * @brief A benchmark evaluating `cudax::cuco::hyperloglog::add_async` performance.
- */
-template <typename Key>
-void hyperloglog_add(nvbench::state& state, nvbench::type_list<Key>)
-{
-  using estimator_type      = cudax::cuco::hyperloglog<Key>;
-  using sketch_size_kb_type = typename estimator_type::sketch_size_kb;
-
-  const auto num_items      = static_cast<cuda::std::size_t>(state.get_int64("NumInputs"));
-  const auto sketch_size_kb = sketch_size_kb_type{static_cast<double>(state.get_int64("SketchSizeKB"))};
-
-  const auto device = cuda::device_ref{0};
-  cuda::stream stream{device};
-  const cuda::device_memory_pool_ref mr = cuda::device_default_memory_pool(device);
-
-  auto items = cuda::make_device_buffer<Key>(stream, device, num_items, cuda::no_init);
-  thrust::sequence(thrust::cuda::par_nosync.on(stream.get()), items.begin(), items.end(), Key{0});
-
-  estimator_type estimator{stream, mr, sketch_size_kb};
-  stream.sync();
-
-  state.add_element_count(num_items);
-  state.add_global_memory_reads<Key>(num_items, "InputSize");
-
-  state.exec(nvbench::exec_tag::timer, [&](nvbench::launch& launch, auto& timer) {
-    timer.start();
-    estimator.add_async({launch.get_stream()}, items.begin(), items.end());
-    timer.stop();
-
-    estimator.clear_async({launch.get_stream()});
-  });
-}
-
-NVBENCH_BENCH_TYPES(hyperloglog_e2e, NVBENCH_TYPE_AXES(bench::defaults::key_type_range))
-  .set_name("hyperloglog_e2e")
-  .set_type_axes_names({"Key"})
-  .add_int64_power_of_two_axis("NumInputs", {30})
-  .add_int64_axis("SketchSizeKB", {8, 16, 32, 64, 128, 256});
-
-NVBENCH_BENCH_TYPES(hyperloglog_add, NVBENCH_TYPE_AXES(bench::defaults::key_type_range))
-  .set_name("hyperloglog_add")
-  .set_type_axes_names({"Key"})
-  .add_int64_power_of_two_axis("NumInputs", {30})
-  .add_int64_axis("SketchSizeKB", {8, 16, 32, 64, 128, 256});