Add a `complete` method for `epi_df`s

[brookslogan]

Currently, we expect users to appropriately tidyr::complete their data appropriately before using with some epiprocess functionality, but complete on epi_dfs drops epi_df-ness; we should stop that from happening. (If we were to enforce key uniqueness, this might need to be done carefully or with some additional checks, but since we don't, it's probably just unconditionally reclassing or dplyr_reconstructing).

[brookslogan]

Adding a complete method may not be the "right" approach. It might naturally fall out of fixing more core functionality like joins. Related: #223.

[brookslogan]

Migrating from #138.

This might be a tool to use in #256; if we want to complete slide inputs, we may need a complete / gap-filling method for that works across both epi_dfs and tibbles (depending on whether our slide inputs have been decayed into tibbles or are still epi_dfs). tidyr::complete may or may not work for this already, but it's probably what we want to use. We're trying to avoid heavy tidyverts dependencies now.

[dshemetov]

Added in #488