Issue 1: Infinite Blocking During Rebalance

The on_partitions_assigned callback attempted to acquire a lock token from SpiceDB via gRPC
without any timeout. When SpiceDB was slow or unavailable:

• gRPC call would hang indefinitely (no timeout)
• Callback blocked for 12+ seconds
• Messages eventually processed WITHOUT a valid lock token
• Violated fencing mechanism designed to prevent split-brain scenarios

Issue 2: Infinite Retry Loop

When messages were processed without a lock token:

• Consumer retried the message indefinitely
• Eventually hit max retries and entered infinite sleep loop
• Pod never crashed, requiring manual intervention
• Retrying was pointless as lock token wouldn't appear without restart