Fixes vmm_struct_cleanup

Cleanup is called from __proc_free, and I don't want to sleep in there.

Running the existing code with the lock debugger will flip out.  Even
attempting to qlock while holding a spinlock will panic, rightly so.
The lock held happens to be the parent's child lock, which it holds when
it decrefs the child.  Regardless, proc_decref is called all over the
place, and I'd rather not sleep in the release method.

I considered removing the qlock and having a multi-state atomic_t that
we CAS on, solving the issue of multiple potential initializers and
cleanup not checking the lock.  But that's a bit of overkill.