select.epi_df falls back to tibble if geo_value or time_value are removed

[dshemetov]

We should either document this explicitly or error.

r$> epiprocess::jhu_csse_daily_subset %>% select(cases)
# A tibble: 4,026 × 1
   cases
 * <dbl>
 1     6
 2     4
 3     6
 4    11
 5    10
 6    18
 7    26
 8    19
 9    23
10    22
# ℹ 4,016 more rows
# ℹ Use `print(n = ...)` to see more rows

[brookslogan]

• dplyr_extending says decaying is the way you should do it, at least.
• And ideally this would mainly happen when the user doesn't need epi_dfness anymore anyway. One exception before was .x fed to a geo-grouped slide computation, but recent slide updates have moved to .keep = TRUE so .x will now be an epi_df there, too.

So seems like explicitly documenting might be the way to go.