Supporting FailMasterPromotionIfSQLThreadNotUpToDate config

[shlomi-noach]

Storyline:

• Avoid promotion if replica not up-to-date with relay log (configurable?) #88
• Slaves lagging by couple of hours are elected as master by orchestrator #83

This PR provides a simple solution to #83 by introducing the FailMasterPromotionIfSQLThreadNotUpToDate boolean config.

When this config is "true", and master failover takes place:

• after the topology is healed
• and candidate replica is put on top
• and just before considering this a success
• and before running hooks
• We verify that SQL thread is up to date with IO thread, effectively meaning all relay logs are consumed.
  • and if that is not the case, i.e. if the SQL thread is not up do date, i.e. some relay log entries are still pending, then we fail the failover and return with error.

To clarify:

• detection+hooks still take place
• topology still healed
• no post-hooks are executed
• failover returns with error.

cc @sougou @adkhare @bbeaudreault @sjmudd

[shlomi-noach]

@sougou @adkhare @bbeaudreault any thoughts on your side?

[sougou]

@shlomi-noach The change looks good to me.