There is a bug here in that snap stop doesn’t obey the timeout set by the snap.yaml.
Specifically, the service-control change creates a exec-command task, which is hard-coded to fail after 61 seconds. See https://github.com/snapcore/snapd/blob/master/overlord/servicestate/servicestate.go#L111
Fixing this however is probably non-trivial and requires most of the work detailed in Systemctl service management unification, which is queued up but not yet being actively worked on.