Apply patience-based early stopping to validation losses and return the best epoch, value, and stopping point.
