Gravar-mail: Sampling from Determinantal Point Processes for Scalable Manifold Learning