[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/python/cuda_cccl/cuda/cccl/headers/__init__.py

@@ -0,0 +1,7 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+from .include_paths import get_include_paths
+
+__all__ = ["__version__", "get_include_paths"]

cccl_upstream/python/cuda_cccl/cuda/cccl/headers/include/__init__.py

@@ -0,0 +1 @@
+# Intentionally empty

cccl_upstream/python/cuda_cccl/cuda/cccl/headers/include_paths.py

@@ -0,0 +1,51 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+import sys
+from dataclasses import dataclass
+from functools import lru_cache
+from importlib.resources import as_file, files
+from pathlib import Path
+from typing import Optional
+
+# type: ignore[import-not-found]
+from cuda.pathfinder import find_nvidia_header_directory
+
+
+@dataclass
+class IncludePaths:
+    cuda: Optional[Path]
+    libcudacxx: Optional[Path]
+    cub: Optional[Path]
+    thrust: Optional[Path]
+
+    def as_tuple(self):
+        # Note: higher-level ... lower-level order:
+        return (self.thrust, self.cub, self.libcudacxx, self.cuda)
+
+
+@lru_cache()
+def get_include_paths(probe_file: str = "cub/version.cuh") -> IncludePaths:
+    cuda_incl = find_nvidia_header_directory("cudart")
+    if cuda_incl is None:
+        raise RuntimeError("Unable to locate CUDA include directory.")
+
+    with as_file(files("cuda.cccl.headers.include")) as f:
+        cccl_incl = Path(f)
+
+    probe_file_path = Path(probe_file)
+    if not (cccl_incl / probe_file_path).exists():
+        for sp in sys.path:
+            cccl_incl = Path(sp).resolve() / "cuda" / "cccl" / "headers" / "include"
+            if (cccl_incl / probe_file_path).exists():
+                break
+        else:
+            raise RuntimeError("Unable to locate CCCL include directory.")
+
+    return IncludePaths(
+        cuda=cuda_incl,
+        libcudacxx=cccl_incl,
+        cub=cccl_incl,
+        thrust=cccl_incl,
+    )