the backoff patroni needed to stop over-eager failovers

*A 200ms network hiccup no longer demotes your primary.*

Patroni used to pull the failover trigger the moment a heartbeat gap appeared, even fleeting ones. @cyberdem0n's #3453 adds a primary race backoff — now a brief network stutter doesn't declare the leader dead. The delay sits in the leader race detection path, buying time before the cluster panics and promotes a replica.

etcd v3's Unavailable exception now lands somewhere safe

*Patroni stops going blind when the store is temporarily unreachable.*

The Unavailable exception from etcd v3 was propagating uncaught, leaving patroni without cluster state visibility during transient store outages. #3562 catches and handles it properly. Now patroni degrades gracefully when the consensus store hiccups instead of crashing blind.

global thread pool now shuts down last

*The shutdown sequence was demoting threads while they were still needed.*

#3569 reorders patroni's shutdown lifecycle — the global thread pool now terminates after everything else has finished. Before, other components tried to schedule work on a pool that was already dead, causing silent failures during shutdown.

patroni now handles threading changes in python 3.11+

*Internal threading behavior shifted; patroni adapts.*

Python 3.11+ changed how the threading module works internally, breaking some of patroni's assumptions about thread lifecycle. #3526 adds compatibility logic to handle the new behavior. Patroni users on newer Python versions no longer encounter threading-related failures.