Avoid selecting servers with excessive replication lag as master

In a two node setup with one master and one slave, if the slave starts to severely lag behind the master in terms of GTID (not seconds_behind_master), the monitor should not promote the slave if the last known GTID coordinate of the master is ahead by some configurable amount.

An alternative approach to this would be to avoid failing over if the monitor is able to detect that transactions would be lost in the failover process. To allow for some amount of losses to happen, the number of transactions that are "an acceptable loss" should be configurable. For example, losing a handful of transactions might be a manually recoverable situation but losing thousands of transactions is not.