Divide integration tests to two clusters

Currently we have test cluster with two storage replicasets. One consists of master and replica while second contains only master. select query is sent to ro instances (replicas), so when there is async replication we can get incorrect sbroad.execute result (see details) and it is not sbroad's problem.

We need two test clusters:

• with master and replica storages for configuration checks (see this MR)
• only with master storages for all other tests