Fix CI workflows for pandas-tests and add test summary.

[bdice]

Description

This PR fixes issues with the pandas-tests job that were introduced during the pandas 2 migration.

It also closes #14846 by adding GitHub Actions summaries for all wheel test jobs, including cudf.pandas. Depends on rapidsai/shared-workflows#173.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[bdice]

/merge

[vyasr]

One thing I'd like to see changed, but this PR generally looks good and I don't think it's worth blocking merge if you don't want to make the change, so resolve as you see fit.

[vyasr]

Did you choose to use --numprocesses instead of -n to match the conda tests? I'm fine with either option but I think (without having done any real investigation) that -n is more common. That may also be a dichotomy between wheels/conda (since we implemented GHA for each in parallel at roughly the same time there was less coordination/enforcement of uniformity) that we could try to unify over time.

[bdice]

We've generally chosen to write the full names of flags wherever possible in our CI scripts. You're correct, this was an artifact of us doing isolated work on conda and wheels at the same time. I took this opportunity to make them match.

[vyasr]

Can we just inline this variable? It's only used once, it's defined far from that location (the pytest call), and it's incongruous with how we provide basically every other argument inline.

[bdice]

It's "far" in lines of code but not really that far semantically. It's just that there's another very large variable definition in between. I think I'm going to leave this as-is because I don't think it is worth another CI run to change.