Gradient Q$(\sigma, \lambda)$: A Unified Algorithm with Function Approximation for Reinforcement Learning