Speed up autoplot.epi_archive

[brookslogan]

With daily-time daily-version data and multiple locations, the wait time can be an annoyance, or it can brick some machines (maybe from memory usage expanding quadratically). Here's some progress toward a faster approach, at least if there aren't a bunch of revisions:

tibble(
  current = s05_smoothed_phrru_archive$DT %>%
    as.data.frame() %>%
    as_tibble(),
  preceding =
    s05_smoothed_phrru_archive$DT[
      unique(s05_smoothed_phrru_archive$DT[, key_colnames(s05_smoothed_phrru_archive), with = FALSE])[
      , time_value := epiprocess:::time_minus_n_steps(time_value, 1L,  s05_smoothed_phrru_archive$time_type)
      ],
      on = key_colnames(s05_smoothed_phrru_archive),
      roll = TRUE
    ] %>%
    as.data.frame() %>%
    as_tibble(),
  subsequent =
    copy(s05_smoothed_phrru_archive$DT)[
    , .version := version
    ][
      unique(s05_smoothed_phrru_archive$DT[, key_colnames(s05_smoothed_phrru_archive), with = FALSE])[
      , time_value := epiprocess:::time_plus_n_steps(time_value, 1L,  s05_smoothed_phrru_archive$time_type)
      ],
      on = key_colnames(s05_smoothed_phrru_archive),
      roll = TRUE
    ] %>%
    as.data.frame() %>%
    as_tibble()
) %>%
  filter(current$geo_value == "103") %>%
  mutate(current_lag = current$version - current$time_value) %>%
  # ggplot(aes(colour = current$version)) %>%
  ggplot(aes(colour = current_lag)) %>%
  `+`(geom_segment(aes(x = preceding$time_value, xend = current$time_value,
                       y = preceding$s05_smoothed, yend = current$s05_smoothed),
                   function(tbl) {
                     tbl %>% filter(!is.na(preceding$s05_smoothed))
                   })) %>%
  `+`(geom_segment(aes(x = current$time_value, xend = subsequent$time_value,
                       y = current$s05_smoothed, yend = subsequent$s05_smoothed),
                   function(tbl) {
                     tbl %>% filter(!is.na(subsequent$s05_smoothed), current$version != subsequent$.version)
                   })) %>%
  # `+`(scale_colour_viridis_c(trans = "date")) %>%
  `+`(scale_colour_viridis_c()) %>%
  {}

todo: faceting, isolated points