On Scalable Testing of Samplers