See #2219 for fuller context.

This is blocked on #2225

• (MVP + 1): We'd like to use the per-sled NTP daemon as the VM's NTP server...
• (MVP): ... But if that's too much, we just need to plumb something - we can point them at the same NTP server used by the whole rack.
• Related; we really just need to be consistent about "which NTP server we use" - they can differ with details like implementation of leap second smearing, which could result in inconsistent time if desynchronized.