Penalty-based Methods for Simple Bilevel Optimization under Hölderian Error Bounds