Is there anything I am missing? In this thread, @bgoodri says
I would say that your priors are putting positive probability on regions of the parameter space with high curvature and / or low numerical accuracy but conditional on the data, those regions have zero probability. So, your priors are probably too weak or otherwise not that consistent with the data, but the data saved you and the posterior draws are fine to make inferences with.
but if anything my priors are pretty narrow are they not? I played around with various widths and nothing makes it really better.