Deep Linear Networks for Matrix Completion -- An Infinite Depth Limit