Currently, when benchRun runs commands and those commands error, we don't really have a way of knowing besides hoping the log debug level is high enough. This caused a perf workload error to go unnoticed.

For bench.cpp specifics, OpType::UPDATE, OpType::INSERT, and OpType::REMOVE  don’t care about whether the command fails or succeeds.

Investigate what type of error handling is necessary here. I imagine certain error codes would be fatal. Over slack, Matt Broadstone suggests wrapping an "ok" check in a MONGO_unlikely (perf impact of this would need to be tested), or implementing a dry run to do error checking.