vm-instance releases report Ready regardless of what the VirtualMachine is doing #4146

packages/system/vm-instance-rd/cozyrds/vm-instance.yaml declares no healthCheckExprs, so a VMInstance release reaches Ready as soon as Helm has applied its objects, regardless of what the VirtualMachine underneath is doing. A VM that never starts, or whose disk never populates, gives its owner a green application.

This is the same shape being closed for vm-disk, where a DataVolume that never binds left the release healthy for as long as it sat there. Disks were the case somebody reported; instances have the same hole and nobody has walked into it loudly yet.

The instrument exists and is already wired: spec.release.healthCheckExprs on the ApplicationDefinition is fed to the HelmRelease at pkg/cmd/server/start.go, and pkg/config/config.go selects the poller wait strategy when exprs are present. What a VMInstance gate needs is the CEL over KubeVirt's own VirtualMachine status rather than over CDI's, so it is a separate piece of work with its own phase vocabulary rather than a copy of the disk one.

Two things the disk work learned that apply here before anyone writes the expressions. A catch-all current swallows the shapes that mean "the controller is not driving this object", so the settled set has to be named rather than defaulted. And a gate on a slow operation needs its install window raised to match, or a healthy-but-slow start reports as a failed release for the rest of its run.

Worth checking the rest of the RD tree in the same pass: vm-disk-rd is currently the only ApplicationDefinition in the tree that sets healthCheckExprs at all, so the question of which other kinds report green while their workload is broken is open beyond these two.