Reconsider handling of locale for step_date()

[juliasilge]

In #1000 step_date() was changed so the locale is recorded at prep() and then enforced at bake(). It turns out this presents problems for deploying a trained model to a different infrastructure. I noticed this when training a model on macOS (where I have en_US.UTF-8) and the trying to deploy on Linux. You can see how the problem plays out here.

That workflow (training locally on macOS, deploying on Linux) is pretty common but there are apparently more complex problems when Windows is involved.

The locale is only used for names of days of the week and months of the year, so is it possible we don't need the full locale system? Can we use clock instead?

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex https://reprex.tidyverse.org) and link to this issue.