On the Decreasing Power of Kernel and Distance based Nonparametric Hypothesis Tests in High Dimensions