[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/examples/stf/1f1b.cu

@@ -1,118 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// 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) 2022-2024 NVIDIA CORPORATION & AFFILIATES.
-//
-//===----------------------------------------------------------------------===//
-
-/**
- * @file
- * @brief Toy example to reproduce the asynchrony of a 1F1B pipeline
- */
-
-#include <cuda/experimental/stf.cuh>
-
-using namespace cuda::experimental::stf;
-
-__global__ void forward(slice<int>, long long int clock_cnt)
-{
-  long long int start_clock  = clock64();
-  long long int clock_offset = 0;
-  while (clock_offset < clock_cnt)
-  {
-    clock_offset = clock64() - start_clock;
-  }
-}
-
-__global__ void backward(slice<int>, long long int clock_cnt)
-{
-  long long int start_clock  = clock64();
-  long long int clock_offset = 0;
-  while (clock_offset < clock_cnt)
-  {
-    clock_offset = clock64() - start_clock;
-  }
-}
-
-int main(int argc, char** argv)
-{
-  context ctx;
-  // Use a graph context if the second argument is set and not null
-  if (argc > 2 && atoi(argv[2]))
-  {
-    ctx = graph_ctx();
-  }
-
-  int device;
-  cudaGetDevice(&device);
-
-  // cudaDevAttrClockRate: Peak clock frequency in kilohertz;
-  int clock_rate;
-  cudaDeviceGetAttribute(&clock_rate, cudaDevAttrClockRate, device);
-
-  auto occ_f = reserved::compute_occupancy(forward);
-  auto occ_b = reserved::compute_occupancy(backward);
-
-  int factor = 1;
-  if (argc > 1)
-  {
-    factor = atoi(argv[1]);
-  }
-
-  size_t num_batches = 8 * factor;
-  int num_devs       = 8;
-  int real_devs;
-  cuda_safe_call(cudaGetDeviceCount(&real_devs));
-
-  std::vector<logical_data<slice<int>>> data;
-
-  for (size_t b = 0; b < num_batches; b++)
-  {
-    auto batch_data = ctx.logical_data(shape_of<slice<int>>(1024));
-    data.push_back(batch_data);
-
-    ctx.task(exec_place::device(0), data[b].write())->*[](cudaStream_t, auto) {
-      // Init ...
-    };
-  }
-
-  cuda_safe_call(cudaStreamSynchronize(ctx.fence()));
-
-  size_t niter = 10;
-
-  for (size_t iter = 0; iter < niter; iter++)
-  {
-    for (size_t b = 0; b < num_batches; b++)
-    {
-      for (int d = 0; d < num_devs; d++)
-      {
-        ctx.task(exec_place::device(d % real_devs), data[b].rw())->*[=](cudaStream_t s, auto bd) {
-          int ms                  = 10;
-          long long int clock_cnt = (long long int) (ms * clock_rate / factor);
-          forward<<<occ_f.min_grid_size, occ_f.block_size, 0, s>>>(bd, clock_cnt);
-        };
-      }
-      //        }
-      //
-      //        for (size_t b = 0; b < num_batches; b++) {
-      for (int d = num_devs; d-- > 0;)
-      {
-        ctx.task(exec_place::device(d % real_devs), data[b].rw())->*[=](cudaStream_t s, auto bd) {
-          int ms                  = 20;
-          long long int clock_cnt = (long long int) (ms * clock_rate / factor);
-          backward<<<occ_b.min_grid_size, occ_b.block_size, 0, s>>>(bd, clock_cnt);
-        };
-      }
-    }
-
-    /* We introduce a fence because the actual pipeline would introduce
-     * some all to all communication to update coefficients */
-    cuda_safe_call(cudaStreamSynchronize(ctx.fence()));
-  }
-
-  ctx.finalize();
-  return 0;
-}