Striving for simplicity and performance in off-policy DRL: Output Normalization and Non-Uniform Sampling