the per-pds account counts drifted from the repo rows on databases written before every row write moved them, and nothing ever repaired them, so a host could stay short or keep counting accounts it no longer has. v13 recounts them from the rows the way v6 does: active rows per host of their pds url, replacing every stored count and pending delta, so a host whose rows all went inactive drops to zero. the counting moves into v6::recount_pds_accounts, which v6 and v13 share with only the row decode differing. host limit statuses stay as they are, since the firehose recomputes them on connect.