Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable