The DNS manager and other singletons have the problem that
they are not properly destroyed on exit, that is, we leak
most of the instances. That should be eventually fixed and
all resources/memory should be released.

Anyway, fix the shutdown procedure by adding an explict command
nm_dns_manager_shutdown(). We should not rely on cleanup actions
to take place when the last reference is dropped, because then
we get complex interactions where we must ensure that everybody
drops the references at the right pointer.

Since the previous shutdown action was effectively never performed,
it is not quite clear what we actually want to do on shutdown.
For now, move the code to nm_dns_manager_stop(). We will see if
that is the desired behavior.