Approximating Bayesian confidence regions in convex inverse problems