Value Activation for Bias Alleviation: Generalized-activated Deep Double Deterministic Policy Gradients