Fix timing issues in flaky scaledown tests

[bjosv]

• Terminate a clusterclient directly instead of first closing its client socket. This avoids that hiredis-cluster might succeed to reconnect to the server before the server has shutdown.

• Close the simulated-redis's listener socket early improves stability.

• Add check for alternative output from clusterclient in testcase cluster-scale-down-test.sh, dbsize-to-all-nodes-during-scaledown-test.shand dbsize-to-all-nodes-during-scaledown-test-async.sh, which depends on timing.

[bjosv]

I'll rerun the checks in an hour, I've seemed to trigger API rate limiter with all my testing. The fixed testcases seems to work better now though..