Beyond the Generative Learning Trilemma: Generative Model Assessment in Data Scarcity Domains