Lemma 3 in the conference version had an error. Specifically, in its proof, on page 16, it is written "E_{\kappa, s_g, s_h, r} [ ...] ... which by the chain rule ... = \frac{2}{n} E_{\kappa, s_h, r}[...]". This is incorrect, because r depends on h, and h depends on s_g. Thus one cannot apply chain rule to remove s_g as done in the proof. We have replaced Lemma 3 with a different, correct statement. Namely rather than obtaining that D' and D are close in KL-divergence, we show that the marginals D'[X] and D[X] are not too far (and similarly for Y), and that the expected conditional distributions of X|Y=y from D' and D (where y is sampled according to D') are close (and similarly for the roles of X and Y reversed). We have adjusted Lemma 5 to interface correctly with this new statement. Our final results are unchanged.