gres count change on reboot

This prevents a slurmctld abort if a job was running on a node
  using all gres/gpu and the node configuration changed to have
  fewer gres/gpu. slurmctld would abort on restart with the new
  configuration.