[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/thrust/testing/min_and_max.cu

@@ -0,0 +1,69 @@
+#include <thrust/extrema.h>
+
+#include <unittest/unittest.h>
+
+template <typename T>
+struct TestMin
+{
+  void operator()()
+  {
+    // 2 < 3
+    T two(2), three(3);
+    ASSERT_EQUAL(two, ::cuda::std::min(two, three));
+    ASSERT_EQUAL(two, ::cuda::std::min(two, three, ::cuda::std::less<T>()));
+
+    ASSERT_EQUAL(two, ::cuda::std::min(three, two));
+    ASSERT_EQUAL(two, ::cuda::std::min(three, two, ::cuda::std::less<T>()));
+
+    ASSERT_EQUAL(three, ::cuda::std::min(two, three, ::cuda::std::greater<T>()));
+    ASSERT_EQUAL(three, ::cuda::std::min(three, two, ::cuda::std::greater<T>()));
+
+    using KV = key_value<T, T>;
+    KV two_and_two(two, two);
+    KV two_and_three(two, three);
+
+    // the first element breaks ties
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::min(two_and_two, two_and_three));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::min(two_and_three, two_and_two));
+
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::min(two_and_two, two_and_three, ::cuda::std::less<KV>()));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::min(two_and_three, two_and_two, ::cuda::std::less<KV>()));
+
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::min(two_and_two, two_and_three, ::cuda::std::greater<KV>()));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::min(two_and_three, two_and_two, ::cuda::std::greater<KV>()));
+  }
+};
+SimpleUnitTest<TestMin, NumericTypes> TestMinInstance;
+
+template <typename T>
+struct TestMax
+{
+  void operator()()
+  {
+    // 2 < 3
+    T two(2), three(3);
+    ASSERT_EQUAL(three, ::cuda::std::max(two, three));
+    ASSERT_EQUAL(three, ::cuda::std::max(two, three, ::cuda::std::less<T>()));
+
+    ASSERT_EQUAL(three, ::cuda::std::max(three, two));
+    ASSERT_EQUAL(three, ::cuda::std::max(three, two, ::cuda::std::less<T>()));
+
+    ASSERT_EQUAL(two, ::cuda::std::max(two, three, ::cuda::std::greater<T>()));
+    ASSERT_EQUAL(two, ::cuda::std::max(three, two, ::cuda::std::greater<T>()));
+
+    using KV = key_value<T, T>;
+    KV two_and_two(two, two);
+    KV two_and_three(two, three);
+
+    // the first element breaks ties
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::max(two_and_two, two_and_three));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::max(two_and_three, two_and_two));
+
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::max(two_and_two, two_and_three, ::cuda::std::less<KV>()));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::max(two_and_three, two_and_two, ::cuda::std::less<KV>()));
+
+    ASSERT_EQUAL_QUIET(two_and_two, ::cuda::std::max(two_and_two, two_and_three, ::cuda::std::greater<KV>()));
+    ASSERT_EQUAL_QUIET(two_and_three, ::cuda::std::max(two_and_three, two_and_two, ::cuda::std::greater<KV>()));
+  }
+};
+SimpleUnitTest<TestMax, NumericTypes> TestMaxInstance;