Test suite: improve wait_until_node_has_status()

Test suite: improve wait_until_node_has_status()

This currently does "onnode any ... wait_until ...".  If ctdbd is
being shutdown on a node then that node might be chosen anyway, if it
is asked early enough.  Then we'll loop on that node but our ctdb
client command may always fail, causing a timeout rather than the
expected behaviour.

This puts the loop on the outside of the "onnode any" so that if the
"wrong" node is chosen initially then on the next iteration the choice
can be remade.

Signed-off-by: Martin Schwenke <martin@meltin.net>