It's possible for operations to be triggered in the background during unit tests that interact with sharding subsystems (e.g. ShardRegistry lookups). If this happens around teardown, these operations will block the executor that they run on. If a mock network is used, it will clear these operations internally, but not necessarily send failures to interrupt waiters. We should explicitly kill all operations during teardown to ensure such operations do not prevent executor shutdown.