There is a race in the 408 test for rabbitmq where the config-change to
enable ssl causes a leader-settings-changed hook in the non-leader units
which results in a rabbitmq service restart. This can happen at exactly
the same time as the test attempts to establish a connection with the
that unit. This patch retries the connection attempt.
Note that this may only be a partial fix as it's possible that a restart
will happen just after the connection is made, which would then result
in a test failure.
Related-Bug: LP#2002156