Currently, errors are published under the general "implements error" interface. We should have a better error publishing strategy such that clients can recognize certain important types of errors and do something intelligent with them.

In addition, it will be important to be able to track root cause errors such that internal implementations can handle certain types of errors better. For instance, when a connection encounters a read error, it should immediately cause the monitor to recheck the server to see if it's down and subsequently scrub the connection pool. This currently isn't happening.