slice_min() differs from dplyr behavior in the presence of NULLs

[MichaelChirico]

conn |>
  tbl(sql("SELECT 1 AS a UNION ALL SELECT NULL as a UNION ALL SELECT 3 AS a")) |>
  slice_min(a, n=2)
# # Source:   SQL [?? x 1]
# # Database: MyConnection
#         a
#   <int64>
# 1      NA
# 2       1

vs.

data.frame(a = c(1, NA, 3)) |>
  slice_min(a, n=2)
#   a
# 1 1
# 2 3

This seems at odds with the documentation where na_rm=TRUE is the intended default. I don't see anything about NULL in the generated SQL, either:

conn |>
  tbl(sql("SELECT 1 AS a UNION ALL SELECT NULL as a UNION ALL SELECT 3 AS a")) |>
  slice_min(a, n=2) |>
  show_query()
# <SQL>
# SELECT `a`
# FROM
#   (
#     SELECT `q01`.*, RANK() OVER (ORDER BY `a`) AS `col01`
#     FROM
#       (SELECT 1 AS a UNION ALL SELECT NULL AS a UNION ALL SELECT 3 AS a) `q01`
#   ) `q01`
# WHERE (`col01` <= 2)