On Statistical Inference with High Dimensional Sparse CCA