Fit operation TestFail needs to be implemented properly

The idea seems to be "if you reach here, you fail the test".  This is a bit tricky in cxx transactions now, since we don't raise exceptions.  Now...

One way to go is jump through some hoops in fit to recognize a previous error (which someone could have noticed and handled).   But, similar to really long lambdas, which timeout part way through, this feels like maybe it is calling out for us to think a bit harder.   Maybe?   

We could return the error, then notice it before continuing in the lambda (through a yet to be designed operation) and then immediately the transaction could be retried or an error returned from run.   Hmm...   Just thinking out loud.   Easier to just be smarter in fit, probably?