Tracebacks may flood

In some circumstances, Swift services (and possibly other daemons) may hit a permanent condition that results in a Python traceback. If that occurs, the master process keeps re-spawning workers and fills the logs. This was demonstrated by e.g. Red Hat bug 967631 "swift keeps spamming console when it cannot access config."

This needs to be fixed somehow. Possible approaches may include:
 - Throttle or rate-limit worker restart, regardless of reason or using the exit codes
 - Split the exceptions into restartable and not-restartable classes, by catching generic things like OSError and KeyError early, then raising SwiftStopError or something