Checking return codes in MadGraphControl

We expected MadGraph would be communicating errors by error messages rather than return codes, which is mostly correct (for MadGraph internal errors, the process still returns 0). What we failed to consider was that sometimes these processes could return an error for some other reason, and if the entire stack isn't killed then the generation job would just happily continue.

In this particular case, MadSpin hit the memory limit and failed, but the rest of the generation job continued happily, because MadGraph had already generated events - the events that MadGraph made were used, rather than the ones after MadSpin, which resulted in incorrect physics output (particularly terrifying).