Add three tests for handling invalid metrics when scaling on
multiple metrics - one for scaling up successfully (new behaviour)
and two for ensuring we don't scale down (existing behaviour).