Add complete.epi_df method so grouped completions don't drop epi_df…

[brookslogan]

…ness

Requesting review from @dshemetov since we talked about S3 stuff recently + to point at #391 where David's already done some of the work we were talking about. @nmdefries I'm not sure this really needs 2 people to review, but GitHub's not letting me unrequest.

[Actually this isn't ready for review yet. I should rebase on top of the time type changes and re-run tests first. And test the time type reinference.]

Checklist

Please:

• Make sure this PR is against "dev", not "main" (unless this is a release
  PR).
• Request a review from one of the current main reviewers:
  brookslogan, nmdefries.
• Makes sure to bump the version number in DESCRIPTION. Always increment
  the patch version number (the third number), unless you are making a
  release PR from dev to main, in which case increment the minor version
  number (the second number).
• Describe changes made in NEWS.md, making sure breaking changes
  (backwards-incompatible changes to the documented interface) are noted.
  Collect the changes under the next release number (e.g. if you are on
  1.7.2, then write your changes under the 1.8 heading).
• See DEVELOPMENT.md for more information on the development
  process.

Change explanations for reviewer

Adds a complete.epi_df method that doesn't drop epi_dfness when grouped.

Too much context:

We've had to basically always recommend complete()ing before putting things into an epi_df, because of various things that could go wrong down the line. While epi_slide_opt() avoids bad behaviors when running on uncomplete()d data, we still aren't free: there is still cmu-delphi/epipredict#362 to avoid and complete()ion for other purposes, such as sliding functions epi_slide_opt() doesn't handle (and epi_slide() does bad things on), filling with 0s, making dplyr::lag actually work, naturally, maybe fixing some other 3rd-party library stuff, and standardizing time availability across epikeys (though we probably actually work for that last case already).

Grouped completions drop epi_dfness; it's annoying to have to carefully place the complete() before as_epi_df() when tweaking epi_df construction (especially if using the new as_epi_df renaming conveniences where you need to remember what the old names were when writing the complete()), and even more annoying/error-inducing to try to convert back to an epi_df after losing epi_dfness (since it's easy to forget and annoying to provide metadata).

Magic GitHub syntax to mark associated Issue(s) as resolved when this is merged into the default branch

• Part of Patches for dplyr type conversions #391; probably the most useful item on the list.

[brookslogan]

Okay, think this is probably ready now.

failed brainstorm, feel free to disregard: I realize now I'm not taking advantage of the time_type attribute to e.g., set a default period, and there may be some overlap with the reasoning used to transform before for epi_slide; we might want to export some helper function that gives an object representing a time difference of 1 period given the time type (such that adding/subtracting it from corresponding time_values "does the right thing"). (or alternatively a function taking an integerish (vector) number of time periods and a time_type, and outputs an object (vector) representing time differences). For the sake of keeping this PR simple, I won't add this unless you think the period default or such a helper function would be a very handy feature. [Nevermind, a period default in complete is a bad idea, because it assumes they want full_seq which they might not. The helper I mention here sounds useful but wouldn't be used in this PR.]