Created by: Haichao-Zhang

This PR add a fill_zeros_array_op to support zero-filling an LoDTensorArray. The dims of input and the output are the same.