The current select mechanism was using the memory too as a consumable
resource, which by default only sets 1 MiB per node. As each job already
requests 1 MiB, it prevents other jobs from running.
As we are not really concerned with memory usage, we only use the unused
cores in the select criteria.
