kernel -- vm_object locking Part 1: Primitives

kernel -- vm_object locking Part 1: Primitives

   * Create per-vm_object 'hold count'; the hold count ensures that a
     vm_object is not deallocated or collapsed till it transitions 1->0.
     Hold count is waited for in vm_object_terminate and _collapse.

   * Convert vm_object_lock/_unlock to use pool tokens; pool tokens do not
     share storage with the vm_object, simplifying lifetime.

Cleanups:

   * Remove first attempts at vm_object locking; it just wasn't clear
     what each token was protecting earlier.

   * Eliminate vestigial fields from vm_object structure.

   * Clean vm_object_pip_wakeup (just call _wakeupn)

   * Remove vm_page_(un)lock for now

Discussed-with: dillon@