A Gradient-free Penalty ADMM for Solving Distributed Convex Optimization Problems with Feasible Set Constraints