Tools for selecting a default evaluation time

[topepo]

show_best() will pick an evaluation time for a dynamic metric when none is given.

Previously, we would find what was in the data and select a time that was close to the median time. This was fine but inconsistent with other parts of tidymodels that do similar operations. For example, tune_bayes has to have a metric to optimize on so it uses the first metric in the metric set and, if needed, the first evaluation time given to the function.

This PR adds a few exported but internal helper functions (primarily first_eval_time()) as the canonical tools for these selections.

This is one of a sequence of PRs:

1. you are here
2. functions to validate the chosen metric and evaluation times
3. refactor the internals of show_best() to be more modular with these tools
4. propagate these changes to other functions that either use show_best() or do similar computations (see When we need to default to a single value for eval_time #766)
5. update finetune to use the change
6. add additional functions that return (potentially) multiple evaluation times (eg autoplot())
7. write an article on tidymodels.org about this.

[hfrick]

There are a couple of suggested changes in terms of style for the tests so highlighting here one comment on structure: We also have logic on when to error/warn in terms of combinations of metric type and eval_time in check_eval_time() which we may want to consolidate in the series of PRs here.

[topepo]

This will pass CI when #776 is merged :-<

[topepo]

Now I remember why check_select_dots() uses enquos(...) instead of list(...)... the latter evaluates the items and that fails when the dots contain something like desc(K) (like in the examples).

[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.