Avoid removing a running driver's PID and socket files if the new driver process fails to initialize #1501
Description
Among remaining issues from #1423 to investigate:
Can we jump through some hoops to keep the original driver instance's PID file in place, if a later driver tried to start and failed to initialize? After all, this later driver process dies quickly and the original blocker remains - but now "hidden" for PIDfile lookups...
Originally posted by @jimklimov in #1423 (comment)