Fit performer error checking bug

In the fit performer, when we get an exception in an operation, we check the type and match against what was expected. That throws a TestFailure when there is a mismatch. However, that runtime_error then is processed by the transactions lambda. This unexpected runtime_error is then dealt with by failing the transaction, after rolling back. Sadly, this can lead to situations where the test appears to pass, if the result was a fail after rollback.

What we need to do is keep track of this failure, and after the lambda returns, we need to then abort the GRPC call. Perhaps. That at least insures the test fails. Perhaps there is a more clever way to accomplish the same thing?

Probably a better way would be to have nested exceptions supported - at least the ones expected to come out of transactions.run(). Then, we just make sure that is set to TestFailure, and check for it and abort.