Reboot a sandbox

Orderly cold-reboot of a running sandbox: the guest shuts its processes down (SIGTERM, then SIGKILL a few seconds later) and flushes its filesystem before the VM is stopped and started again from the same disk on the same node. In-memory state and running processes are lost, but recently written files survive; the disk and IP are preserved. If the guest doesn't respond within a few seconds it falls back to a hard reset, in which case un-synced writes can be lost. A pending resize applies on reboot. For an unconditional hard reset, use reset-sandbox.