Revisiting Optimal Convergence Rate for Smooth and Non-convex Stochastic Decentralized Optimization