Implement n_distinct() for the Arrow Substrait compiler

[thisisnic]

n_distinct() was implemented in #244 for DuckDB, but cannot yet be implemented for Arrow as it's not been implemented in libarrow to be translated to the appropriate Acero function.

Here's a possible implementation a test:

arrow_funs[["n_distinct"]] <- function(x, na.rm = FALSE) {
  check_na_rm(na.rm)
  warn_n_distinct()

  substrait_call_agg(
    "aggregate_approx.approx_count_distinct",
    x,
    .output_type = substrait_i64(),
    .phase = 3L,
    .invocation = 1L
  )
}

test_that("arrow translation for n_distinct works", {
  skip_if_not(has_arrow_with_substrait())
  skip("Substrait function `approx_count_distinct` not yet implemented in Arrow C++")

  expect_equal(
    tibble::tibble(x = c(1:5, 1:3)) %>%
      arrow_substrait_compiler() %>%
      substrait_aggregate(n = n_distinct(x, na.rm = TRUE)) %>%
      dplyr::collect(),
    tibble::tibble(
      n = 5
    )
  )
})