fix/better persistent leader unloading

[stanley2058]

This PR changed how the persistent leader works to handle edge cases better:

1. If a server is a persistent leader for a namespace, it will now extend the cleanup timeout until there are no new messages from Redis.
2. After the persistent leader is unloaded after the final persistent, it will immediately remove the leader key, indicating there's no leader for the namespace in the cluster.