Supervisor: with the Runner CRD backend, a restore that fails on its
node is detected and reported with the reason, instead of only surfacing
when the platform's heartbeat timeout fires.
Mono-RevId: c95a9ebc98d5b336717fba5cae03510a46c5b19c