Remove ZK watch for completing partition reassignment

This is a follow-on from KAFKA-8836.

Currently we have a ZK watch on the partition "/state" znode which fires a handler in the controller to check if a reassignment can be completed due to replicas catching up with the leader. This is located in KafkaController#processPartitionReassignmentIsrChange

Following the change to updating ISR with the new AlterIsr RPC, we would like to remove this ZK watch and replace it with a direct call when writing out a new ISR in the controller.