single_builds records r_version per attempt, but the skip query ignored
it and matched on platform and arch alone. A non-primary pass therefore
skipped every package the primary pass had already attempted under a
different minor: build-all.R --sensitive-only under R 4.6 skipped
packages that had only ever been built for 4.5, so the per-minor slots
never filled and the backlog could not be worked off by rebuilding.
That is why amd64/resolute serves a 4.6 client 22322 packages against
the 4.5 slot's 26346, and why the run intended to close that gap
reported 'Skipped 2334 already-attempted package versions; 59 remaining'.
Matching on the major.minor prefix rather than the full string keeps a
patch bump (4.6.0 -> 4.6.1) from re-attempting the whole catalogue.