Currently if do_next_deploy/clean_step fails, the failure is swallowed
by eventlet since they're run in a new thread. This 1) is incorrect,
2) leads to nodes stuck in DEPLOYING/CLEANING.
Also update logging in agent_base to be able to easier spot similar
problems.
Change-Id: I0282c9e06c54a173efc666cd8df25cf573afb394