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