cluster_SUITE: Use peer's standard_io to communicate with spawned nodes

[Why]
When running test cases in parallel and thus starting and stopping
Erlang nodes frequently, we sometimes get an error from `global` that
breaks the corresponding test case. That's because Erlang distribution
tries to connect every nodes together by default.

[How]
By using peer's standard_io communication, we ensure started nodes don't
try to form a full mesh network and interfere with each others.

Author: dumbbell