Another way is to get just the first row for each consecutive label. We can do that before the call to plot, or one way that can work better at times (especially if ggplot() is after a sequence/pipe of data-changing commands) is to change data= for a particular geom using its rlang-~ faux-function interface.

BTW, it's better to not use dt$ within ggplot aesthetics.

Working code:

ggplot(dt, aes(x = year, y = value)) +
  geomline() +
  scaleycontinuous(labels = dt$label, breaks = dt$value) + 
  geomtext_repel(
    # removed dt$ from dt$label
    aes(label = label),
    # filter/change the data ggplot knows, effective for just this geom
    data = ~ .x[, .SD[1,], .(label, rleid(label))]
  )

The .x of data = ~ .x[...] is replaced with the core data that ggplot "knows" within the plot-space.

(Small caveat: the use of rleid assumes that the data is ordered within the frame. Easy to work through if that is not guaranteed, either by forcing the order or perhaps some other techniques.)

ugly solution:

ref %
  mutate(lagLabel = lag(label)) %>%
  mutate(changeLabel = ifelse(label == lagLabel, NA, label))
g