LiteASR: Efficient Automatic Speech Recognition with Low-Rank Approximation