Prototype based Masked Audio Model for Self-Supervised Learning of Sound Event Detection