Add expected performance checks in "TestOrchestration" tests

[AlanRostem]

After discovering issue #142, it would be nice to compute the expected number of executed commands, failures, and timeouts for each test based on the configuration.

Since the tests are done on local replicas, the computation of expected values should be trivial and somewhat deterministic, given some margin of error.

[meling]

I've been looking at the metrics package, and it appears suitable for recording such values (executed commands, failures, and timeouts) so that they can be used to fail tests that are clearly outside the normal range. We should also add counter metrics for weird stuff, like block already exists, vote too old, and other weirdness.