Generic multi-thread/cross-process error handling framework

The idea for a generic inter-process error-handling framework came from working on HBASE-6055 (snapshots). Distributed snapshots require tight time constraints in taking a snapshot to minimize offline time in face of errors. However, we often need to coordinate errors between processes and the current Abortable framework is not sufficiently flexible to handle the multitude of situations that can occur when coordinating between all region servers, the master and zookeeper. Using this framework error handling for snapshots was a simple matter, amounting to maybe 200 LOC.

This seems to be a generally useful framework and can be used to easily add inter-process error handling in HBase. The most obvious immediate usage is as part of HBASE-5487 when coordinating multiple sub-tasks.