Before this commit, in Sentinel mode, when the already known sentinel instance switch address, there is a chance making same sentinel instance being considered as two separate sentinel instances which may causing problem.
For example in the following scenario:

1. we are deploying three sentinels (sentinel1: 127.0.0.1:26379, sentinel2: 127.0.0.1:26380,sentinel3:127.0.0.1:26381)monitoring same redis master/replica instance
2. we shutdown sentinel 1 down for maintainance, and we start a new sentinel 4 (127.0.0.1:26379) from scrach, which will generate a new runid, in sentinel 2 and sentinel 3's view, when it get the hello package from sentinel 4, sentinel1 port number will be set to 0 temporatorily, but not been deleted.
3. after sometime we shutdown sentinel 4 and make sentinel 1 back online, in that case, sentinel1 and sentinel 4 will both exist in sentinel 2/3's list, with same ip and port, and this causing the duplication issue. In sentinel 2/3's point of view the sentinel 4 is even ONLINE since the health check (ping/pong) package doesnt contain the runid.

The root cause of this issue is when removing sentinel with same runid, we still need to invalidate other known sentinels with same ip/port.
https://github.com/redis/redis/blob/unstable/src/sentinel.c#L2874