Hi,

While working on some NUMA code recently, I stumbled over how nodemask
operations are unconditionally using MAX_NUMNODES length, which is typically
1024 for distro configs.

OTOH typical machines only have <=4 nodes, so doing these crazy long bitmap ops
is silly.

Therefore do as cpumask does and use nr_node_ids analogous to nr_cpu_ids.

These patches are lightly tested -- as in they seem to boot on a 2 node system.