On a Reinforcement Learning Methodology for Epidemic Control, with application to COVID-19