Fix race condition on batch job termination that could result in a job exit
code of 0xfffffffe if the slurmd on node zero registers its active jobs at
the same time that slurmstepd is recording the job's exit code.
but 535