Copyright on this BIND 10 work from CZ.NOC belonfs to ISC.

(reviewed on jabber)

This branch removed the shared pointer that the NSAS used to link
to the resolver.  As a result we now have to make sure that the
resolver stays in existence while the NSAS is active.  This commit
fixes that problem in a unit test.

If so, for now treat these as a timeout and resend if there are retries left, fail otherwise

Conflicts:
	src/lib/resolve/recursive_query.cc

Conflicts:
	ChangeLog
	configure.ac
	src/bin/bind10/tests/bind10_test.py.in

.stop() followed directly by .reset() doesn't work. It seems asio has a
"can run" boolean variable, stop sets it to false, reset to true, so if
they are both called from inside the .run() main loop, it does nothing.

Without this it cannot use privileged port (most commonly 53) while changing
the run time user (-u).  Confirmed the fix, and okayed on jabber.

I won't update changelog for this fix because it's a regression after
the previous release.

Reverted ZoneEntry to the original version and added the destructor
to TestResolver to call all the saved callbacks.  This breaks
internal shared_ptr loops, allowing the NSAS to destoy itself
gracefully, and avoiding memory leaks.

Previously when compiled without NDEBUG set it did, if a check
failed it called assert().  The tests checks for this using
ASSERT_DEATH.  However, ASSERT_DEATH leaks memory and this was
obscuring the valgrind output.  The changes here cause an exception
to be thrown (instead of a call to abort()) in the case of an error
and the unit tests now EXPECT_THROW instead of ASSERT_DEATH.

The ZoneEntry object was creating a callback and keeping a shared
pointer to it, but the callback was also using a shared pointer to
link back to the ZoneEntry.  This change removed the shared pointers
in the callback objects - they now use a "raw" pointer to point back
to the ZoneEntry.