kgs-genmove_cleanup, UCT pass_all_alive: Fix up semantics
Dan Reish pointed out that Pachi mishandles kgs-genmove_cleanup;
it turns out KGS misbehaves wrt. dead stones in case Pachi agrees to
stop the game during kgs-genmove_cleanup, but wms does not want
to fix that.
So, make Pachi strictly follow the genmove_cleanup protocol - instead
of passing only if Pachi wins when all stones are marked alive, pass
if no stones are considered dead. This is less friendly to the user,
but strictly by the protocol.