Keep epi_df-ness in summarise + fix roxygen typos

[brookslogan]

No description provided.

[brookslogan]

Closes #296. But this may not be the best way to do it. Maybe we'd get some of the intended behaviors of these reconstructing wrappers via implementing group_data.epi_df. But some might be unavoidable depending on grouped_df method implementations. There's also still issues with the metadata when binding rows which might be easier to resolve in some wrappers like the current approach (though none of them do that right now). Especially with dplyr 1.1.0 summarize that deprecates returning != 1 row per group, we may also have some re-keying hints that we might need to act upon.

[brookslogan]

I think an alternative approach involving an "grouped_edf" class is going to be preferable.