The Optimal Approximation Factors in Misspecified Off-Policy Value Function Estimation