Different RedisFailover's sentinels join together

[dbackeus]

I'm reopening what was probably the most critical bug before the fork: spotahome#550

It can result in complete data loss for a cluster. Not just in theory, it happened in our production k8s cluster. Eg. if cluster A gets mixed up with cluster B and cluster A ends up being elected master for all nodes, the previously existing data of cluster B will get completely overwritten by cluster A's.

I still believe the appropriate solution is to start relying on hostnames instead of injecting pod IP's into config as mentioned in spotahome#550 (comment)