Optimal asymptotic MSE of kernel regression estimate for continuous time processes with missing at random response