Optimal Designs for Random Coefficient Regression Models with Heteroscedastic Errors