[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/libcudacxx/include/cuda/__cmath/isqrt.h

@@ -0,0 +1,80 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of libcu++, the C++ Standard Library for your entire system,
+// 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) 2025 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _CUDA___CMATH_ISQRT_H
+#define _CUDA___CMATH_ISQRT_H
+
+#include <cuda/std/detail/__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
+
+#include <cuda/std/__bit/integral.h>
+#include <cuda/std/__concepts/concept_macros.h>
+#include <cuda/std/__type_traits/is_integer.h>
+#include <cuda/std/__type_traits/is_signed.h>
+#include <cuda/std/__type_traits/make_unsigned.h>
+
+#include <cuda/std/__cccl/prologue.h>
+
+_CCCL_BEGIN_NAMESPACE_CUDA
+
+//! @brief Returns the square root of the given non-negative integer rounded down
+//! @param __v The input number
+//! @pre \p __v must be an integer type
+//! @pre \p __v must be non-negative
+//! @return The square root of \p __v rounded down
+//! @warning If \p __v is negative, the behavior is undefined
+_CCCL_TEMPLATE(class _Tp)
+_CCCL_REQUIRES(::cuda::std::__cccl_is_integer_v<_Tp>)
+[[nodiscard]] _CCCL_API constexpr _Tp isqrt(_Tp __v) noexcept
+{
+  if constexpr (::cuda::std::is_signed_v<_Tp>)
+  {
+    _CCCL_ASSERT(__v >= _Tp{0}, "cuda::isqrt requires non-negative input");
+  }
+
+  if (__v <= 1)
+  {
+    return __v;
+  }
+
+  using _Up = ::cuda::std::make_unsigned_t<_Tp>;
+
+  _Up __uv = static_cast<_Up>(__v);
+  _Up __ret{};
+  _Up __bit = static_cast<_Up>(_Up{1} << ((::cuda::std::bit_width(__uv) - 1) & (~1)));
+
+  while (__bit != 0)
+  {
+    if (__uv >= __ret + __bit)
+    {
+      __uv -= __ret + __bit;
+      __ret = (__ret >> 1) + __bit;
+    }
+    else
+    {
+      __ret >>= 1;
+    }
+    __bit >>= 2;
+  }
+  return static_cast<_Tp>(__ret);
+}
+
+_CCCL_END_NAMESPACE_CUDA
+
+#include <cuda/std/__cccl/epilogue.h>
+
+#endif // _CUDA___CMATH_ISQRT_H