Better handling of failures

Until now we didn't take good care of how Barista fails. We have some tests to check that it fails, but not how it fails.

After looking around in the code I found out some problems with failures: Sometimes we notify Espresso's failure handler, some times we don't (#220), sometimes we throw a BaristaException with a nice message, sometimes we just throw the default Espresso AssertionError, sometimes the error messages don't give any hint about the real issue.

It's a mess.

I'm going to try to fix that incrementally. I'm opening this issue to serve as a common reference for the PRs I will open.