Sparsity-Aware Distributed Learning for Gaussian Processes with Linear Multiple Kernel