Sentinel discovery traffic always lands on the same sentinel node, causing load imbalance

Version: redis-py 8.1.0

Platform: Python 3.13 on Debian Trixie

Description:

When using Sentinel, all master/replica discovery traffic seems to consistently land on the same sentinel node rather than being distributed across the configured sentinels. In practice this means one node ends up carrying noticeably more load than the others, which sit mostly idle.

This isn't a new observation — it was reported before in #1179, and there were two earlier attempts to address it (#654, #770), both of which stalled and were closed as stale without real maintainer discussion. Since the underlying behavior hasn't changed, I wanted to resurface it in case there's renewed interest in revisiting it, especially for setups running many sentinel-connected clients where the imbalance becomes more noticeable over time.