pf_unique_count, pf_unique and pf_rank -- questions about repeated values.
All three rest on one engine call (engine_build_runs), which sorts and reports where the runs
of EQUAL rows are in the same pass. Equality is the sort comparator's own, minus the index
tiebreaker that makes it a total order -- so "distinct" here means exactly "the sort would not
have to choose between them", and two NaNs are one value even though == says otherwise.
Nulls are outside the population, in all three. They are excluded from a count, excluded
from the distinct values, and given rank 0 rather than a place in the ranking. That is why none
of the three takes nulls_first: there is no null tier to position. Extraction is therefore
always nulls_first=.false., which puts every null last and contiguous -- the property the
walks below rely on to stop counting.