Mining Order-preserving Submatrices under Data Uncertainty: A Possible-world Approach and Efficient Approximation Methods