Change the method to get the launch

Currently, the most recent launch is used to retrieve the computed value, after checking the state of the firework. A more reliable method would be to use fireworks.Workflow._get_representative_launch() to get the launch. This method is hidden (underscored), uses a firework object and returns a launch object. These objects can be large. Maybe re-implement the same algorithm in an utility function and use only small projections?