This problem was reported by Harvard University and could be
reproduced with a command line of "srun -N1 --tasks-per-node=2 -O id".
With other job types, the error message could be logged many times
for each job. This change logs the error once per job and only if
the job request does not include the -O/--overcommit option.