Change affinity so that cores from different numa nodes are selected

in the cpuid map. These are now offered in numa node order rather
than core order. This means that hyperthreads should be picked last.

With this fix we look to work well with Intel MPI using multiple ranks
on the same workstation, but PLATFORM MPI remains uncooperative.
To help with that a new command-line option -a has been added to
disable processor affinity, even when the numa policy is set
and available.