I think the OP was hell-bent on implementing exactly the JAGS model, so I avoided suggesting he do something else. But in general I 100% agree a different prior would be better. My impression is that many papers choose certain priors because they are convenient for the sampler the authors were working with (often Gibbs). The same considerations that lead them to choose these priors are often either not releveant or actively harmful when switching to HMC.