orc-agent signal handling

[Igneous]

Currently, if orc-agent is in the middle of a handler run (like chef + chef-client), it will respond to a SIGTERM by terminating and sending SIGTERM to the processes forked from it.

We need a "graceful stop/restart" mechanism, so sending orc-agent a SIGTERM causes a check for currently running handlers. If there are currently running handlers, wait until the run finishes, then terminate.

Another thought: should we unbind from our queues and de-register our amqp consumers, so we don't recieve any more messages after a SIGTERM?