JVM: Use worker pool for RPC handler selection and catch errors internally

[daviddawson]

Currently, breakpointing in the handler selection will cause all event dispatch to freeze, generally leading to transport disconnection/ keep alive channel failure.

This can be avoided if the stack switched to using its pool straight away rather than selecting the handler first.

Also, errors from handler selection/ processing need to be trapped more effectively and logged with request info to give debugging info.