Implement +maffinity support using hwloc

[stwhite91]

The +maffinity option is potentially useful for running processes across NUMA domains. The Phoronix NAMD benchmark uses this option with a multicore build to run with a single process across an entire node. Currently there are two issues with this:

1. The +maffinity option is disabled/ignored in the NAMD release binaries because Charm++ was built without libnuma.

2. Instead of relying directly on libnuma, Charm++ should use hwloc for implementing memory affinity more portably. Currently Charm++ only uses hwloc to implement cpu / thread affinity, but it supports memory affinity as well: https://hwloc.readthedocs.io/en/stable/group__hwlocality__membinding.html