Polynomial Escape-Time from Saddle Points in Distributed Non-convex Optimization