epiprocess doc cleanup

[brookslogan]

• Fix non-linked vignette("backtesting", package="epipredict") in archive vignette; downlit should be linking this if it exists? downlit might be unable to link because we're not on CRAN yet.
• "Data frame returns will be tidyr::unpack()-ed, if named, and will be tidyr::pack-ed columns, if not named."
  • This is the opposite of what happens; we pack only if they are "named" using .new_col_name =. Readers may also confuse "named" with "has column names", which data.frames should/must have, or maybe even row names.
• Link/use to deaths-from-cases&deaths forecasting example instead of current bad examples.
  • Think this was already addressed (Fix prob_arx misalignment in this repo as well, or remove buggy vignette( section)s #100).
• Document NA-filling of input windows in epi_slide.
• Fix mix of [.]version[s] and [.]ref_time_value[s] in ?epix_slide. (We might mention you can use .ref_time_value as well as .version inside formulas and data-masking expressions, but we should point to .version first in this case, and to version/.versions in all other discussion.)
• Make ?epi[x]_slide actually describe what they do. (Starting paragraph maybe isn't really that clear.)
• [-] Try to remove some unnecessary details/comparison in ?epix_slide
  • Deferring to https://github.com/cmu-delphi/forecasting-team-meta/issues/29
• Remove link to removed "advanced" vignette.
• Remove bad epix_slide example
• [-] Remove overly-complicated / less relevant vignettes.
  • Deferring to https://github.com/cmu-delphi/forecasting-team-meta/issues/29
• Fix garbled .all_versions docs
• Fix Details: second bullet point formatting regarding .all_versions
• [-] Separate out new_* functions into separate documentation pages?
  • Deferring to https://github.com/cmu-delphi/forecasting-team-meta/issues/29
• Fix misleading/confusing ?epi_df "Slides a given function over variables in an epi_df object.`"
• Replace median example in ?epi_df with something that's not dominated by epi_slide_opt(frollapply).
• NEWS: add non-technical main highlights
• Describe NA behavior noted below.

Thanks @XuedaShen for identifying some of these.

[XuedaShen]

Also a reminder that epi_slide completes the input window with NA, thereby leading to downstream NAs if we don't manually specify na.rm. I wasn't able to find a remark on either complete or NA in the docs of epi_slide.

[dajmcdon]

• General practice: all functions should have examples to illustrate at least 1-2 intended use cases.
• epi_slide_opt() Description field should say "when and why to use this and not epi_slide()" And probably port back to the other. Would be nice to have examples to illustrate the improvement (or cases not handled here).
• In complete.epi_df() rather than saying "see tidyr::complete" for each argument, it should use #' @inheritParams tidyr::complete. The Description field should say what it does. Then add a link to the tidyr version in the See also field.

[brookslogan]

I believe we have a lot more "see {x}" than just in complete.epi_df. We can't blindly apply @inheritParams for all of these cases; I think I found some case where ... might have been misleading. Anyway, we just need to read the results of trying @inheritParams and sanity-check / override when necessary.

[One reason we might have used "as in" is to highlight what actually changed compared to some other function. We should probably decide on this tradeoff case by case.]