`n()` ignores ordering in window_frame

[mattcane]

When using n() in an ordered window frame the ordering seems to be ignored when it's translated to SQL. Use case would be knowing how many rows are actually included in a rolling sum/average/etc.

mf <- memdb_frame(
  tibble(
    grp = c(rep('a',4), rep('b', 5)),
    order = seq(1,9),
    val = seq(11,19)
  )
)

mf %>%
  group_by(
    grp
  ) %>%
  # Arrange by order
  window_order(order) %>%
  # Calc rolling 2 row sum
  window_frame(-2,-1) %>%
  mutate(
    roll_last_2_sum = sum(val),
    # Get number of rows in frame
    n_rows_in_frame = n()
  ) %>%
  show_query()

# Output is:
# SELECT `grp`, `order`, `val`, SUM(`val`) OVER (PARTITION BY `grp` ORDER BY `order` ROWS BETWEEN 2 PRECEDING AND 1 PRECEDING) AS `roll_last_2_sum`, COUNT(*) OVER (PARTITION BY `grp`) AS `n_rows_in_frame`

# Output should be:
# SELECT `grp`, `order`, `val`, SUM(`val`) OVER (PARTITION BY `grp` ORDER BY `order` ROWS BETWEEN 2 PRECEDING AND 1 PRECEDING) AS `roll_last_2_sum`, COUNT(*) OVER (PARTITION BY `grp` ORDER BY `order` ROWS BETWEEN 2 PRECEDING AND 1 PRECEDING) AS `n_rows_in_frame`