This improves/fixes a couple of things:

It's possible for the server to reply with a 207 (multi-status) code,
which implies that one or more events in the batch might have been
improperly formatted. Let's handle this condition in the same way we
handle actual errors (400 and above).

In addition, when sending events, the EventProcessor only catches
IOExceptions, when in fact there might be other types of exceptions that
come from the EventSender. If an uncaught exception reaches the Executor
that is running the sending task, the executor might not run the task
again, causing events to stop being sent altogether for the remainder
of the VM lifecycle. Even though catching all exceptions is a Bad Thing,
in this rare case catching everything is actually beneficial and will
help us track down problems easier.