Toward better NUMA scheduling

The second system call, in particular, is useful for something like QEMU if you're emulating a guest that's large enough to have multiple virtual NUMA nodes. It let's QEMU tell the kernel the relevant areas of the guest's memory that corresponds to the virtual NUMA nodes without getting into the business of doing explicit pinning.

Userspace really has no business doing CPU or memory pinning IMHO so I'm glad to see the kernel start to be more proactive here.