[CI] Add CPU-only CI job(s)

[jameslamb]

Description

This project should add one or more CI jobs running some subset of the tests in an environment without a GPU.

Benefits of this work

Improves release confidence by reducing the risk of merging changes that break CPU-only support.

Approach

This could be accomplished by using the custom-job shared action from shared-workflows (code link) and targeting a CPU node, like this:

  cpu-only-tests:
    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@branch-24.06
    with:
        node_type: cpu4
        build_type: pull-request
        run_script: "ci/run-cpu-tests.sh"

For an example, see how cudf runs the pandas unit tests on a CPU-only runner:

https://github.com/rapidsai/cudf/blob/516d0f9033e73d10a473e2ca3fcc891e980450bc/.github/workflows/pr.yaml#L182-L189

Notes

This issue is inspired by a recent experience over on ucx-py. There, while testing rapidsai/ucx-py#1041, a CPU-only job (in that example, docs building) caught that some changes would have made that library unusable without access to a GPU.

For more details, see rapidsai/ucx-py#1041 (comment).

[luweizheng]

Hi @jameslamb

Is there a CPU-only pypi or conda package of ucxx. I used to install by conda install -c conda-forge -c rapidsai ucx-proc=*=cpu ucx ucx-py, and now I am going to move to ucxx.
I now can install by pip install ucxx-cu12 on a CPU node, and it seems that most of ucxx's functions work, but it has some warning of not having cuda devices. What's more, I am not sure if the performance of the GPU-compiled package is fast enough and whether I need to compile a CPU version.

If there is a CPU version wheel or conda package, it would be great.

[pentschev]

Since this is issue is about adding CPU-only CI jobs and not packages, they could be tackled separately, would you mind filing a new issue with that request @luweizheng ?