Flexible Bayesian methods for cancer phase I clinical trials. Dose escalation with overdose control