investigate replacing most of the producer machinery with yet-another-logger

[jonsterling]

Then, we may just export a logger backend for yet-another-logger. It might let us delete a lot of code, and have a single place where we implement and test the super gnarly error handling logic that has to occur for a logger.