[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/test/support/fp_compare.h

@@ -0,0 +1,58 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, 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
+//
+//===----------------------------------------------------------------------===//
+#ifndef SUPPORT_FP_COMPARE_H
+#define SUPPORT_FP_COMPARE_H
+
+#include <cuda/std/algorithm> // for cuda::std::max
+#include <cuda/std/cassert>
+#include <cuda/std/cmath> // for cuda::std::abs
+
+#include "test_macros.h"
+// See
+// https://www.boost.org/doc/libs/1_70_0/libs/test/doc/html/boost_test/testing_tools/extended_comparison/floating_point/floating_points_comparison_theory.html
+
+template <typename T>
+TEST_FUNC bool fptest_close(T val, T expected, T eps)
+{
+  constexpr T zero = T(0);
+  assert(eps >= zero);
+
+  //	Handle the zero cases
+  if (eps == zero)
+  {
+    return val == expected;
+  }
+  if (val == zero)
+  {
+    return cuda::std::abs(expected) <= eps;
+  }
+  if (expected == zero)
+  {
+    return cuda::std::abs(val) <= eps;
+  }
+
+  return cuda::std::abs(val - expected) < eps && cuda::std::abs(val - expected) / cuda::std::abs(val) < eps;
+}
+
+template <typename T>
+TEST_FUNC bool fptest_close_pct(T val, T expected, T percent)
+{
+  constexpr T zero = T(0);
+  assert(percent >= zero);
+
+  //	Handle the zero cases
+  if (percent == zero)
+  {
+    return val == expected;
+  }
+  T eps = (percent / T(100)) * cuda::std::max(cuda::std::abs(val), cuda::std::abs(expected));
+
+  return fptest_close(val, expected, eps);
+}
+
+#endif // SUPPORT_FP_COMPARE_H