Feature Request: Prefer longer running replicas for PRS

[GuptaManan100]

Feature Description

When PRS-ing currently if all the replicas are caught up, we semi-randomly select one of them. This is non-optimal for large instances, we want to reparent to the replica that has been up for the longest (i.e. has had more time to warm its buffer pool from a possible innodb buffer pool dump upon shutdown). In a typical deployment scenario, the replica with longer uptime might have been running a minute or two longer than the other eligible replica. This isn't much, but that is enough time to load up 5-10 GB or more (depending on volume IOPs) of the buffer pool than the replica with shorter uptime. This enables the replica with longer uptime to immediately perform (potentially significantly) better than the other replica if it is promoted to primary.

Use Case(s)

Better performance from the new primary selected after PRS.