it doesn't make sense in practice and should better be cleaned up.

test cases were adjusted.  in particular, test_do_axfrin_nozone_sqlite3
was simply removed; it's okay as it only intended to the old behavior
itself.

from this point, xfrin stops auto-create a new zone; the admin must explicitly
create empty zone using b10-loadzone -e.

this is a backward compatible change, and is currently no-op in effect, but
will be crucial with main changes of this branch that are coming.

it'll soon be impossible to assume a specific data source for the very
initial transfer, we need to extract the zone creation part from b10-xfrin
(which should be the right course anyway).  the new loadzone option is
my proposed way to do this.  I think this option itself is of some use, too.

we're going to use stored client retrieved from a client list.  at that point
we need to ensure zone finders are only used by a single thread (updaters can
run concurrently by multiple threads).

no behavior change.

this is already unused and has been effectively dead code.

so we can test much larger part of _cc_setup().  right now no behavior change.

Fixed Conflicts:
	ChangeLog

it now unconditionally uses AXFR, so mentioning possible failure of IXFR
doesn't make sense.

actually refresh wasn't defined in spec, so is defined now.
retransfer is revised so it always uses AXFR (BIND 9 compatible).
lettuce tests are adjusted accordingly.

it should now be safe as we (have already) support fallback and use AXFR
initially.

this is essentially a refactoring: there's no behavior change.
but it'll help later part of this branch.

also updated the test so the MockCC refers to the spec default, instead of
hardcoding the values.

no behavior change yet.  also, use AXFR by default, for now, to preserve
the behavior.