Decide on column ordering invariants / lack thereof

[brookslogan]

Related to #166. We should decide whether/where we want to enforce column orderings for epi_dfs and epi_archives, and make sure to account for other_keys. We'd want to resolve/consider some of these points:

• new_epi_df reorders based on geo_value and time_value only
• differences between new_epi_df and as_epi_df noted in Document the differences for column re-ordering in new_epi_df and as_epi_df #166, and dplyr_reconstruct.epi_df not reordering
• small informal poll from Make endpoint arg order consistent: geo, time epidatr#26 supporting geo-time-otherkeys order (and conflict with some archive examples/tests that use geo-otherkeys-time-version ordering)

[brookslogan]

With some of the *canonical functions we've moved epi_dfs to often be geo-otherkeys-time, as that seems more natural for data structures than the geo-time-otherkeys from the API interface poll.

Not sure if this is fully addressed by the *canonical work. There may at least be lingering documentation issues still. Leaving open for now.