Number and the categorical kinds are naturally kind-isolated by the stats machinery already,
with no partition and no new index required: getAggregates is computed over the numeric
population regardless of what else shares the range, and getFrequencies({ filterFn })
restricts distinctness to one kind. Each line leads with the count, exactly like the bare
Number: N a Types section already prints — this is that line, with the stats appended.
Date is left out here, on cost rather than ability: computing it correctly means a value scan
(see countDates) rather than a getCounts/getStats read, and this payload is the
CHEAP alternative to full per-column profiling — a Contentful region listing that pays for a
date scan on every unprofiled row has lost the reason it exists. appendKinds, called once per
report rather than once per listed region, affords it.
Per-kind stats, pooled across the WHOLE range regardless of column — "what's actually in here, broken down by type" in one shot, as opposed to profileSeries's per-column breakdown.