more flexible updatepages() interface
PTE locking (using one of the free PTE bits)
guard against concurrent unmaps (except no way to check if vma is deleted yet..)

CPU may have interrupts disabled waiting for that spinlock too, and
will never acknowledge our tlb flush..

read the next pointer, and then unlock a node.  otherwise, the next pointer
changes and we try to unlock subsequent nodes that we never locked.

 - new crange api
 - fix several crange bugs (see comments)
 - better scoped_acquire
compiles but still buggy and missing tlb shootdowns