Investigate approaches to automatically launch RM eval job after UM training job

[dustinvtran]

@dustinvtran This will also let us automatically run postprocessing-style models such as deep ensembles or greedy HP ensembles.

https://github.com/google/edward2/issues/327