Optimize the per-document cluster state checks in the geoip and ip_location processors

[joegallo]

This is a continuation of #120824

• We can use the threadpool's clocks rather than System.currentTimeInMillis() to get the time more efficiently.
• We can use a better cluster state listener to:
  • watch for changes to the database_validity not-quite-Setting
  • update the map of database files
• Also we currently call checkDatabases on every single cluster state change, that seems excessive.

[elasticsearchmachine]

Pinging @elastic/es-data-management (Team:Data Management)