axis_canvas and ggplot hlines with same values are not aligned when scale_y_continuous uses expand argument

[martindurian]

ggplot v3.3.0
cowplot v1.0

when ggplot uses scale_y_continuous with expand argument, equal values on plot and axis_canvas are not aligned.
when I remove the expand argument, then everything is nicely aligned

library(tidyverse)
library(cowplot)

# dataframe -----
plt_df <- tribble(
  ~Week_Day, ~Week_Day_Name, ~workday, ~Week_day_consumption,
  1, "Mon", TRUE, 23376,
  2, "Tue", TRUE, 23843,
  3, "Wed", TRUE, 20790,
  4, "Thu", TRUE, 20993,
  5, "Fri", TRUE, 18963,
  6, "Sat", FALSE, 15387,
  7, "Sun", FALSE, 17997
)

# calculate averages for week parts -----
avg_df <- summarise(group_by(plt_df, workday), part_mean = mean(Week_day_consumption))

base_plt <- ggplot(plt_df, aes(x = Week_Day, y = Week_day_consumption)) +
  geom_col(aes(fill = workday)) +
  geom_hline(data = avg_df, aes(yintercept = part_mean, color = workday)) +
  theme(legend.position = "none") +
  labs(title = "without expansion")

# create base plot with expansion -----
base_plt_exp <- base_plt +
  labs(title = "with expansion") + 
  scale_y_continuous(expand = expansion(mult = c(0.05, 0.1)))

# alternative y axis plot annotate average lines -----
annotation_plt <- axis_canvas(base_plt,
  axis = "y",
  data = avg_df
) +
  geom_hline(aes(yintercept = part_mean))

annotation_plt_exp <- axis_canvas(base_plt_exp,
  axis = "y",
  data = avg_df
) +
  geom_hline(aes(yintercept = part_mean))

# insert alternative y axis ------
plot_composed <- ggdraw(insert_yaxis_grob(base_plt, annotation_plt, grid::unit(0.1, "null"),
  position = "right"
))

plot_composed_exp <- ggdraw(insert_yaxis_grob(base_plt_exp, annotation_plt_exp, grid::unit(0.1, "null"),
  position = "right"
))

plot_grid(plot_composed, plot_composed_exp)