Created by: Haichao-Zhang

This PR extends the fill_zeros_like_op to support zero-filling an LoDTensorArray. The dims of input and the output are the same.