Probably don't want to drop the grouping in [.epi_df when as_tibble() is used

[rachlobay]

As per [.epi_df, when we subset an epi_df so that the result is not a true epi_df anymore (ex. missing time_value or geo_value), then we return a tibble of the results (refer to around lines 102-104 of methods-epi_df.R). That seems ok, but a problem that arises is that if there is a grouping in the original epi_df, that grouping is lost when we use as_tibble(). This issue showed up in the last two uses of expect_identical() in tests-epi_slide.R. Ex. In

edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01")+200L) %>% 
                     dplyr::select("geo_value","slide_value_value")

the grouping by geo_value is lost when we use dplyr::select("geo_value","slide_value_value").

Currently, we have decided to leave as is, but we may decide in the future that we should change this so the grouping is kept when [.epi_df is used.

Tagged this as a question to keep it open for discussion.

[brookslogan]

Interacts with #223. With the grouped_edf approach, implementing the alternative behavior might be done via implementing some of ?dplyr_extending for grouped_edfs.