Divergence, even just 1, could be an indication of bias estimation. While the rest of the diagnosis all look good, you should do a pairplot and check whether the divergence is systematic or not. If so, more informative prior usually helps; if not, increase the target accept and see if you can get rid of the divergences.