Currently, we detect if we are on a numa machine by examining /sys/devices, and detect how numa is configured by looking at /proc/self/numa_maps. We then issue a warning.

A better way to do this, on machines new enough to offer it, is to use libnuma to query the numa state. Potentially, we could also change the interleave state as well with numa_set_interleave_mask.

This would both protect us from future changes to /proc and /sys, and give us finer grained control over our NUMA story.