Resume paused instances on restore requests

[sjmiller609]

Summary

• Resume the existing VMM when a restore request finds a paused instance, under the existing instance lock.
• Publish the running state and restore event after success; invalidate cached state on resume failure so another wake can retry.
• Leave standby rollback, startup/shutdown, start requests, snapshots, and networking unchanged.

---

Note

Medium Risk
Changes VM lifecycle restore behavior and hypervisor resume/caching under lock; mistakes could leave paused VMs stuck or double-resume, though tests cover concurrency and retry paths.

Overview
Restore requests no longer fail when the instance is paused with an existing VMM. Instead of falling through to standby restore (which rejects non-standby states), RestoreInstance now resumes the live hypervisor under the per-instance lock, refreshes cached hypervisor state, and emits LifecycleEventRestore on success.

On resume failure, cached hypervisor state is invalidated so a later restore can retry. Running and initializing instances remain a no-op, and the full standby snapshot restore path is unchanged.

New tests cover resume success for running/initializing metadata targets, unchanged on-disk metadata, failed-then-success retry, idempotent second restores, and concurrent restores invoking Resume only once.

^{Reviewed by Cursor Bugbot for commit 2653831. Bugbot is set up for automated code reviews on this repo. Configure here.}