Number of client connections not reevaluated dynamically in the teardown script

[jkhalack]

Description

The CONN_COUNT variable is only evaluated once in the zookeeperTeardown.sh script (https://github.com/pravega/zookeeper-operator/blob/master/docker/bin/zookeeperTeardown.sh#L23-L32)
As a result, the loop would always take 30 seconds if there were client connections present when the script started to run, and ZK pod would terminate with 137 error code with the default termination grace period set to 30s (the same as in #91)

Importance

The failure makes it impossible to replace a failed node on K8s cluster (the affected ZK pod would not get shut down properly, hence not possible to migrate it).

Location

https://github.com/pravega/zookeeper-operator/blob/master/docker/bin/zookeeperTeardown.sh#L23-L32

Suggestions for an improvement

Move line 23 (evaluation of CONN_COUNT) inside the for loop, so it gets reevaluated every cycle and allows the code to break out of the loop earlier (then we might also have less time spent in upgrades, as in #206).