Assessing the accuracy of Bayesian-optimized CGMD in predicting polymer miscibility