Atomic Drop Instance can exceed jutemaxbuffer #3073
Description
Describe the bug
Atomic drop creates a transaction to atomically delete all znodes under instance and config path. If this transaction exceeds the jute max buffer, then the client will disconnect from ZK and reconnect under the same session, but the operation will never succeed.
To Reproduce
In our case, the node had thousands of current and customized states.
Expected behavior
Deletion should either be achieved non-atomically or a proper exception should be thrown
Additional context
Atomic delete was implemented as we were facing recurrent issue of instance znodes being partially deleted, which lead to cluster getting into a bad state. Node could only be removed through manual intervention in ZK
#2994