[cuda.compute]: add CI job for minimal cuda-cccl extra

ci/matrix.yaml

@@ -84,6 +84,7 @@ workflows:
     - {jobs: ['test'], project: 'python', ctk: ['12.X',        '13.X'], py_version: ['3.10'], gpu: 'l4', cxx: ['gcc13', 'msvc2022']}
     - {jobs: ['test'], project: 'python', ctk: ['12.X','13.0', '13.X'], py_version: ['3.14'], gpu: 'l4', cxx: ['gcc13', 'msvc2022']}
     - {jobs: ['test'], project: 'python', py_version: '3.14', gpu: 'h100', cxx: 'gcc13'}
+    - {jobs: ['test_py_compute_minimal'], project: 'python', ctk: '13.X', py_version: '3.14', gpu: 'l4', cxx: 'gcc13'}
     # CCCL packaging:
     - {jobs: ['test'], project: 'packaging', ctk: '12.0', cxx: ['gcc10', 'clang14'], gpu: 'rtx2080', args: '-min-cmake'}
     - {jobs: ['test'], project: 'packaging', ctk: '12.X', cxx: ['gcc10', 'clang14'], gpu: 'rtx2080'}
@@ -539,6 +540,7 @@ jobs:
   test_py_headers:  { name: "Test cuda.cccl.headers",      gpu: true,  needs: 'build_py_wheel', force_producer_ctk: "pybuild", invoke: { prefix: 'test_cuda_cccl_headers'} }
   test_py_coop:     { name: "Test cuda.coop._experimental",  gpu: true,  needs: 'build_py_wheel', force_producer_ctk: "pybuild", invoke: { prefix: 'test_cuda_coop'} }
   test_py_par:      { name: "Test cuda.compute",     gpu: true,  needs: 'build_py_wheel', force_producer_ctk: "pybuild", invoke: { prefix: 'test_cuda_compute'} }
+  test_py_compute_minimal: { name: "Test cuda.compute minimal", gpu: true, needs: 'build_py_wheel', force_producer_ctk: "pybuild", invoke: { prefix: 'test_cuda_compute_minimal'} }
   test_py_examples: { name: "Test cuda.cccl.examples",     gpu: true,  needs: 'build_py_wheel', force_producer_ctk: "pybuild", invoke: { prefix: 'test_cuda_cccl_examples'} }
 
   # Run jobs for 'target' project (ci/util/build_and_test_targets.sh):

ci/test_cuda_compute_minimal_python.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+ci_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+repo_root="$(cd "$ci_dir/.." && pwd)"
+source "$ci_dir/pyenv_helper.sh"
+
+# Parse common arguments
+source "$ci_dir/util/python/common_arg_parser.sh"
+parse_python_args "$@"
+require_py_version "Usage: $0 -py-version <python_version>"
+
+cuda_major_version=$(nvcc --version | grep release | awk '{print $6}' | tr -d ',' | cut -d '.' -f 1 | cut -d 'V' -f 2)
+
+# Setup Python environment
+setup_python_env "${py_version}"
+
+# Fetch or build the cuda_cccl wheel:
+if [[ -n "${GITHUB_ACTIONS:-}" ]]; then
+  wheel_artifact_name=$("$ci_dir/util/workflow/get_wheel_artifact_name.sh")
+  "$ci_dir/util/artifacts/download.sh" "${wheel_artifact_name}" "${repo_root}/"
+  wheelhouse_dir="${repo_root}/wheelhouse"
+else
+  "$ci_dir/build_cuda_cccl_python.sh" -py-version "${py_version}"
+  wheelhouse_dir="${repo_root}/wheelhouse"
+fi
+
+# Install cuda_cccl with the minimal CUDA extra. This intentionally avoids the
+# full cu* extras because those pull in numba/numba-cuda.
+CUDA_CCCL_WHEEL_PATH="$(ls "${wheelhouse_dir}"/cuda_cccl-*.whl)"
+python -m pip install "${CUDA_CCCL_WHEEL_PATH}[minimal-cu${cuda_major_version}]"
+python -m pip install pytest pytest-xdist "cupy-cuda${cuda_major_version}x"
+
+cd "${repo_root}/python/cuda_cccl/tests/"
+python -m pytest -n 6 -v compute/test_no_numba.py