feat: CCCL CachingDeviceAllocator preload — 完整依赖链 288 files

从 cccl_upstream 递归追踪 cub/util_allocator.cuh 的全部 include 依赖:
  cub/         9 files (config, util_*, version, detect_cuda_runtime)
  cuda/        libcudacxx type_traits, concepts, algorithm, iterator...
  nv/          target macros, preprocessor

总计 288 个头文件 (1.4MB)，打包到 include/ 目录，编译时 -I include
即可完全脱离 CCCL 原始目录结构。

.cu 文件直接 #include <cub/util_allocator.cuh>，
走原版 CUB CachingDeviceAllocator，零 mock。

BI-V100 参数: growth=2 bins=[8..32] max_cached=8GB/device

qwen3_6_scripts/cccl_preload/include/cub/detail/detect_cuda_runtime.cuh

@@ -0,0 +1,45 @@
+// SPDX-FileCopyrightText: Copyright (c) 2011, Duane Merrill. All rights reserved.
+// SPDX-FileCopyrightText: Copyright (c) 2011-2021, NVIDIA CORPORATION. All rights reserved.
+// SPDX-License-Identifier: BSD-3
+
+/**
+ * @file
+ * Utilities for CUDA dynamic parallelism.
+ */
+
+#pragma once
+
+// We cannot use `cub/config.cuh` here due to circular dependencies
+#include <cuda/__cccl_config>
+
+#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
+
+#ifdef _CCCL_DOXYGEN_INVOKED // Only parse this during doxygen passes:
+//! Defined if RDC is enabled and CUB_DISABLE_CDP is not defined.
+//! Deprecated [Since 3.2]
+#  define CUB_RDC_ENABLED
+
+//! If defined, support for device-side usage of CUB is disabled.
+//! Deprecated [Since 3.2]. Use CCCL_DISABLE_CDP instead.
+#  define CUB_DISABLE_CDP
+
+//! Execution space for functions that use the CUDA runtime API, e.g. to launch kernels. Such functions are `__host__
+//! __device__` when compiling with RDC, otherwise only `__host__`.
+//! Deprecated [Since 3.2]
+#  define CUB_RUNTIME_FUNCTION
+#else // Non-doxygen pass:
+
+#  if _CCCL_HAS_CDP()
+#    define CUB_RDC_ENABLED
+#  endif // _CCCL_HAS_CDP()
+
+#  ifndef CUB_RUNTIME_FUNCTION
+#    define CUB_RUNTIME_FUNCTION _CCCL_CDP_API
+#  endif // CUB_RUNTIME_FUNCTION predefined
+#endif // Do not document