Retrying host maintenance falsely marks already-migrated VMs as stopped

CLOUDSTACK VERSION
4.22.1.1, KVM hypervisor

SUMMARY
Putting a host into maintenance moves its VMs off one at a time, which
can take a while. Requesting maintenance again on the same host before
the first request finishes causes it to retry moving every VM still
listed against that host, including ones already moved. That retry
fails as expected, but for some VMs it confuses CloudStack into
thinking the VM was powered off, and it gets marked stopped even though
it was never touched and kept running the whole time.

EXPECTED
Requesting maintenance again should not retry VMs already moved.

ACTUAL
Some already-moved VMs get falsely marked stopped.

NOTE
The false "stopped" status seems to come from CloudStack losing track
of which VMs are on which host. That could plausibly happen from causes
other than this one, but we only have evidence of it via this scenario.

Start by reproducing repeated maintenance requests on a KVM host in CloudStack 4.22.1.1, then trace the host-maintenance VM migration flow and how the host's VM list is updated. Done means a second request skips VMs already migrated, and those VMs remain running rather than being marked stopped.

Written by the indexing model from the issue text.