Review and consider applying `?dplyr_extending` for `epi_df`

[brookslogan]

I believe that we implemented epi_df without the use of ?dplyr_extending. We may have:

• Missing S3 implementations
• Unnecessary S3 implementations
• Noncompliant behavior

We should review ?dplyr_extending to see if it hints at bugs, enhancements, code simplifications, etc. (E.g., the mention of [ might have hinted that we would want a [ implementation before we discovered it on our own.)

[brookslogan]

Might have ran into one of these cases in #64 where grouped epi_dfs being bound together lost their epi_dfness. Implementing a dplyr_reconstruct.epi_df based on [.epi_df seems to partially resolve. But note this exacerbates #242 (which underlies #213).

May also related to #117, cmu-delphi/epipredict#51.

[brookslogan]

A potential issue with ?dplyr_extending guidance: arrange on grouped epi_dfs drops epi_dfness. We need to implement dplyr_row_slice as well. (We also need to implement other methods interacting with grouping: group_by, ungroup, group_modify [or maybe not; an appropriate group_data + any other required extensions of grouped_df methods (see below) might get this "for free"/"properly"]; as well as unnest; we may be missing some others.)

[brookslogan]

There are a few other issues with grouped epi dfs in the current dev / lcb/make_epix_slide_more_like_reframe branches; see #113, #298. Note:

1. Part of this may be due to missing a group_data method, alluded to in ?dplyr_extending. We'd want either/both (a) "epi_df" before "grouped_df" in the class vector to be able to dispatch to group_data.epi_df instead of group_data.grouped_df, or (b) to mess with the class of the "groups" attr generated by group_by (doesn't sound great.)
2. Note in comments of Make print.epi_df header mirror style of print.tsibble and print.tbl_df #113 that we want "grouped_df" before "epi_df" due to some pillar methods for grouped_df expecting it to be before more base-like classes in the class vector.

1(a) and 2 (and maybe more points) seem to conflict. One potential resolution:

• Make grouped epi_dfs use the class c("grouped_edf", "grouped_df", "epi_df", "tbl_df", "tbl", "data.frame"). That's also suggested in this older thread.

[brookslogan]

While applying the "potential resolution" above seems to help in some aspects, it also loses epi_df-ness in a lot of other situations. The problem appears to be that grouped_df's methods don't really respect dplyr_extending for grouped_* classes in front of it. So we'd need to look at methods(,"grouped_df") and/or methods(,"grouped_{somethingelse}") and make sure to override/mirror the things we want. [2024-09-24: perhaps this wasn't/isn't the case; it seems you can get these ops to work as long as you make sure not to lose your head class after any NextMethods in dplyr_extending impls, including e.g. names<-, and this may be sufficient. Also, looking at grouped_df S3 methods might not have even worked, as there might be some non-S3 functions manually branching on grouped_df-ness.] Related: #145. [We'll need to look in some packages beyond dplyr for these, e.g., for vctrs::restore.]

[Same approach would need to be applied for rowwise edfs. Plus, in any approach, we will want to review tidyr for more methods that may need wrapped like, e.g., unnest.]