feat(virtual-machine): make the guest shutdown grace period configurable - #3991

What this PR does

terminationGracePeriodSeconds is hardcoded to 30 in
packages/apps/vm-instance/templates/vm.yaml, so no consumer of the aggregated API can change how long a
guest is given to shut down cleanly. This exposes it as a chart value, default 30 — behaviour unchanged
on upgrade.

Why 30 seconds is not always enough

Stopping a VM is an ACPI powerdown: the guest closes its applications, flushes its journals, then hands
back control. Linux does that in a few seconds. Windows applies pending updates during shutdown — the "Do
not turn off your computer" screen — which routinely exceeds 30 s, and at 30 s KubeVirt destroys the
domain. Every stop of a machine with queued updates is then an unclean shutdown of the guest filesystem.

And it is not limited to a user pressing stop. The same field bounds every shutdown path: a node
drain, an eviction, and a workload update rolled out by KubeVirt itself (evictionStrategy: LiveMigrate,
workloadUpdateMethods: [LiveMigrate, Evict] on the KubeVirt CR). A cluster upgrade therefore hard-kills
those guests, and no dashboard button is involved to work around it. That is also why the value belongs in
the CR rather than in a stop subresource call: only the CR is honoured by the paths nobody triggers.

What this changes

• packages/apps/vm-instance/values.yaml — annotated @param {int} terminationGracePeriodSeconds, default
  30, with the Windows case spelled out in the comment.
• packages/apps/vm-instance/templates/vm.yaml — renders the value instead of the constant.
• Generated by make generate and committed: values.schema.json, README.md,
  api/apps/v1alpha1/vminstance/types.go, and packages/system/vm-instance-rd/cozyrds/vm-instance.yaml.

Testing

helm template on both paths — value absent renders 30, --set terminationGracePeriodSeconds=300
renders 300.

No helm unittest suite is added for templates/vm.yaml: its instanceType and its disks are gated by
cluster lookup calls, so the template cannot render without a live cluster. That is the same reason the
existing suite in this package scopes itself to templates/vm-update-hook.yaml, and says so in its header.

Screenshots

N/A — no UI change.

Downstream repositories

• No downstream repository is affected by this change
• cozystack/website - follow-up:
• cozystack/terraform-provider-cozystack - follow-up: VMInstance: model the new terminationGracePeriodSeconds field (upstream cozystack#3991) terraform-provider-cozystack#36
• cozystack/ansible-cozystack - follow-up:
• cozystack/ccp - follow-up:
• cozystack/talm - follow-up:
• cozystack/cozyhr - follow-up:
• cozystack/cozy-proxy - follow-up:
• cozystack/cozystack-telemetry-server - follow-up:
• cozystack/external-apps-example - follow-up:
• cozystack/examples - follow-up:

Downstream: walked the trigger map against the diff, file by file.

• cozystack/terraform-provider-cozystack — affected: this adds a field to an app's
  values.schema.json, and the provider models vm-instance config by hand (schema, model, expand/flatten
  pair) in both the resource and the data source. Follow-up issue: cozystack/terraform-provider-cozystack#36
  (searched first: [core] Update fluxcd and kubeapps #35 exists but is scoped to disks[].hotpluggable, a different field).
• cozystack/website — covered by the release bot: vm-instance is in the VMS list of the website
  Makefile, and its reference page is regenerated from this package's README.md, which make generate
  updated here.
• ansible-cozystack, ccp, talm, cozyhr, cozy-proxy, external-apps-example, examples,
  cozystack-telemetry-server — no trigger: nothing under hack/, no package layout, namespace, CRD,
  installer value, variant, node prerequisite or release-prep change.

Release note

feat(virtual-machine): `terminationGracePeriodSeconds` is now a `vm-instance` value (default 30). It bounds the ACPI powerdown of the guest on every shutdown path — an explicit stop, a node drain, an eviction, a workload update — after which KubeVirt destroys the domain. Raise it for Windows guests, which apply pending updates during shutdown and are otherwise hard-killed at 30 seconds.