ACL mode check configurability and optimizations

The first idea here would be to make the ACL mode check min and max intervals configurable.

The second idea would be to optimize the whole ACL mode transitioning process to listen for serf member events instead of periodically checking all members to see if they support new ACLs.

With the second idea there would be no reason for the first as it would be based on serf events and not timers. However its not clear whether we really need to do any of this. Hopefully the transition period is so small that it makes it not worth it.

The main case where this would be useful is in clusters with heterogeneous server versions. If the primary DC is running <1.4.0 but secondary DCs are running 1.4.0+. Those secondary DCs would forever check if the primary DC has been upgraded. Another scenario is when a DCs servers have not been upgraded but the clients have. In that case the clients are going to perform the periodic check to determine when the servers are capable of using new ACLs.

More investigation is needed.