Distributed Matrix Computations with Low-weight Encodings