Finally, add support for the Critical Page Pool to the Slab Allocator.  We
need the slab allocator to be at least marginally aware of the existence of
critical pages, or else we leave open the possibility of non-critical slab
allocations stealing objects from 'critical' slabs.  We add a separate,
node-unspecific list to kmem_cache_t called slabs_crit.  We keep all
partial and full critical slabs on this list.  We don't keep empty critical
slabs around, in the interest of giving this memory back to the VM ASAP in
what is typically a high memory pressure situation.

-Matt