Make bench.h/cpp's handle expected errors and error counts.

Currently all the many uses of stats.errCount in bench.cpp, like this one, don't check error codes. If a test expects certain errors to occur, there should be a way to state what errors are expected to more safely handle errors.