Refactored functions to select single evaluation times

[topepo]

Primarily exports choose_eval_time() for use in select/show best functions.

More changes will be coming for autoplot() and similar.

Places where select_best() is used: augment.tune_results(), fit_best(). These will need more work since they do not have eval_time arguments.

There is also still some legacy code that overlaps with this and previous changes (e.g. middle_eval_time()). These will be removed in a subsequent PR.

1. Tools for selecting a default evaluation time #768
2. Refactor choose_metric #777
3. (you are here) functions to validate the chosen evaluation time with updates to show_best() and select_best() (see When we need to default to a single value for eval_time #766).
4. update finetune to use the change
5. add additional functions that return (potentially) multiple evaluation times (eg autoplot())
6. refactor metric input checking code in the tune_* functions, fit_resamples(). etc.
7. write an article on tidymodels.org about this.

[hfrick]

🚂

[github-actions]

This pull request 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.