Add error handling for async produce

[rnowling-memphis]

The current implementation for asynchronous production of messages doesn't provide an error handling mechanism. We could provide a callback function through the add_done_callback() method to the tasks that checks the return status and either calls a user provided callback function OR adds the message and error information to a list that the user can check and process.

As part of this, we might to distinguish between recoverable and non-recoverable errors. If we hit a non-recoverable error, we might want to stop producing more messages.