Provide a command that summarizes what daemon versions are present, and add a health check that reports if potentially-dangerous mismatched versions are present.

This should *not* fire health errors during normal upgrades.  The condition for health errors is probably to be based on daemon start order: mixtures are acceptable as long as we only see newer-versioned daemons starting afer older versioned daemons.  We might also have a time threshold beyond which mixed clusters are considered unacceptable.