Consider adding fetch functions returning epiprocess objects

[brookslogan]

To streamline analyses using our new package ecosystem, it might help to be able to get epi_dfs directly out of epidatr. This could be opt-in with Suggests: epiprocess and requiring it inside fetch_edf. However, at minimum, this would involve renaming and setting key columns on a per-endpoint basis, and we might also consider pivoting to a wider format. Thus, it's a nontrivial amount of work. (But also hopefully a nontrivial amount of savings when actually being used.)

Might conflict with the goal of #72.

[dshemetov]

I'm thinking we should circle back to this after both #72 is addressed and the v1.0.0 release. A few thoughts on how this might work:

• this would just consume the output of the new fetch interface from Consolidate fetch_* interfaces #72, which will return either an unstructured list (for classic only endpoints) or a tibble
• the tibbles that have the requisite columns to make an epi_df, we can support this interface and otherwise give an error
  • having to write endpoint-specific branches in fetch_edf doesn't sound ideal; checking for requisite columns might get complicated if the endpoints have different names for requisite cols

[brookslogan]

Just realized we might also want to get epi_archives as well. Adding also a fetch_ea() / fetch_archive() / etc. might be too much. Not sure if what we actually want is a fetch_wide(), which would perform pivoting and maybe renaming. One trouble is "issue" vs. "version" naming, and the presence of the "lag" column interfering with epi_archive compactification. (These details actually seem like they start to justify that extra fetch_ea() / function...)

[dshemetov]

Adding fetch_ea seems like a medium difficulty task. Each endpoint would need careful attention to how their particulars (namely, the differences in key columns) are translated to an epi_archive.

[brookslogan]

epi_df also has other_keys metadata, though not essential for construction or some operations. (Though we plan to make it a little more prominent than it is now, e.g., by printing it when we print an epi_df, and making it an explicit parameter of the new_epi_df and as_epi_df functions.) I think epi_archive's key may need to rename issue to version, eliminate lag, and add on version to the epi_df keys.

[brookslogan]

It might actually require more thinking about how to handle issue and lag in epi_df, since (a) we need/want to change them in epi_archive and might want some type of consistency, and (b) these columns are more like values if you do a latest or as_of query, but one or the other is like a key when you do issues or lags queries.

[dshemetov]

Two separate but related points:

1. how about we do provide tibble -> epi_df, epi_archive in epiprocess instead of here?
2. how about we provide recipes instead of full on conversion functions?

Advantages:

• (1) and (2) keep the fetch interface in this client minimal and focused around getting a standard R tibble of data
• (1) separates epiprocess concerns out of this package; allows epiprocess to focus on delivering a tibble -> epi_df pipeline
• (2) concerned that there are too many degrees of freedom in this conversion; not wrapping tibble -> epi_df functions allows us to not worry about constructing an interface and the user can see the inner workings and modify as they see fit