Rework the naModLock()/naModUnlock() semantics to be more

Rework the naModLock()/naModUnlock() semantics to be more
maintainable.  The rules are simple (extension functions are called
*with* the lock, which must be dropped before calling naCall(), which
grabs it) but the tracking of when the lock was held was getting a
little confused.  Keep a "nasal call depth" count in the subsystem to
figure out whether we are making a sub-call and thus hold the lock.