Instead of `puts`ing when the failure occurs save it until the error
message and print a prose description of what the failure was and the
output from the command. This makes the output from failing tests
significantly easier to read.