- every judge and release case of update_marker_corpus.json through the
production marker.ps1 functions (live table, own pid/ct, now injected)
- real processes: claim decided inside the lock; four concurrent
claimants over a dead marker -> one owner; old packaged Desktop cmd.exe
wrapper lineage (start /b and waiting wrapper) adopted, unrelated live
owner refused; -HandoffRun adoption; -MarkerOp reclaim/withdraw/busy;
release hand-over to a live delegate; heartbeat; script killed before
the delegate is published runs no update and leaves a dead marker
(cherry picked from commit b641f0855ceb7f3242edfd0f35c296004b7a95e0)