Try to fix #1157 by using both checks suggested in the issue, namely:

1. We should be prepared to wait longer for the the Kafka controller to delete a topic, since deleting a large number of topics is known to basically lock up the controller for a time. So we bump the timeout for that from 2 to 5 minutes.
2. When we observe the deletion of a topic via ZK don't assume the topic deletion has propagated to observations of the topic via the Admin client. Instead wait for the Admin client to confirm that the topic doesn't exist before proceeding with the reconciliation.