GH-37804: [R] Fix with_language test helper

[nealrichardson]

Rationale for this change

Proper fix for #37804. Turns out the with_language() helper was wrong, but usually in an innocuous way that would make it skip more than it should.

Are these changes tested?

Yes, though we'll want to watch the nightlies and may need to add another nightly that uses rocker/r-devel in order to ensure (such that we can) that whatever changed on CRAN is covered. It's not clear that the R-hub job we test that uses R devel is getting updated nightly anymore.

Are there any user-facing changes?

No.

• Closes: [R] Tests which use with_language() fail on CRAN #37804

[thisisnic]

Now this is being really weird and passing on Windows (and passes on MacOS) but failing on Ubuntu 🙃

[github-actions]

⚠️ GitHub issue #37810 has been automatically assigned in GitHub to PR creator.

[github-actions]

⚠️ GitHub issue #37804 has been automatically assigned in GitHub to PR creator.

[thisisnic]

I'm wondering if this could be relevant here? r-lib/withr#213

[nealrichardson]

Maybe? Note that the with_language() here is not the same as withr::with_language(), and we explicitly look for the case where setting LANG does not result in changing the translation and skip the test if not. It's possible that extra check is wrong--it seemed to be wrong before this PR, but may be wrong differently?