Build runners restart in a loop

Runners on both clusters restarted repeatedly. They reported themselves as running and exited cleanly each time, so it did not look like a crash.

They were several versions behind the platform they connect to. Their health check reads the result of asking for work, that exchange fails across such a version gap, and the check then reports the runner unhealthy and restarts it.

Every runner on both clusters now matches the platform version. One of them had been pinned to nothing at all, so it silently froze on whatever version was current when it was last installed; it now names its version explicitly.