How many DISTINCT non-null values values holds. n_null optionally reports how many
were null.
Nulls are excluded from the population, not counted as one value -- the same rule
pf_nth_quantile follows, and the reason this takes neither descending (a count does
not depend on direction) nor nulls_first (there is no null tier to place).
Distinctness is the sort comparator's own equality, so on a floating-point array it is
EXACT: 0.1 + 0.2 and 0.3 are two distinct values. Every NaN counts as one value,
collectively, since NaNs compare equal to each other here (they do not under ==).