Improve error handling to reduce pipeline stalls

[kkleemola]

Unhandled exceptions cause the runner to retry messages infinitely when in streaming mode. If the error is deterministic (i.e. because it's from parsing a message), the pipeline will stall requiring manual intervention. These types of errors should be caught and outputted (either as a log message we can alert on or as an error message with the original event so we can track down problems easier).

[kkleemola]

Partially solved with #553