Shared 'heavy' slot wrapper: timeout counts queue wait, wrap timeout inside

On a box where heavy jobs go through a flock-based slot wrapper (heavy cmd loops until a slot is free), timeout 3000 heavy cmd starts the clock while the job is still waiting in the queue. When the queue is saturated the job is killed before it runs and leaves an empty log, which looks like a crash. Use heavy timeout 3000 cmd so the limit applies only to the run. Also launch long runs with nohup and append an exit $? line to the log, so a later check can tell 'never ran' from 'still running' from 'finished'.