Step failure doesn't always lead to build failure

[MatthewHambley]

It seems that failure of some steps doesn't always lead to the build stopping. In particular, when analysis failed due to issue #221 Fab carried on to compile which obviously failed.

[bblay]

There's a ticket to revisit the exception handling. I don't like the way we return exceptions from child processes instead of raising them like normal Python. I think if we switch from Pool.map to Pool.map_async we could make use of the error callback and greatly improve the error handling pattern and simplify the code.