* Excluded separate balancers lock from cluster
* Refactored `cluster.Cluster` interface (`Insert` and `Remove` returning nothing now)
* Replaced unsafe `cluster.close` boolean flag to `cluster.done` chan for listening close event
* Added internal checker `cluster.isClosed()` for check cluster state
* Extracted getting available conn from balancer to internal helper `cluster.get` (called inside `cluster.Get` as last effort)
* Added checking `conn.Conn` availability with `conn.Ping()` in prefer nodeID case