Thank you for the answers,
This was very helpful!
First, I did not mention the divergent transitions issue, so thanks for pointing out what it was driven by.
Second, mea culpa for overlooking truncated random number generation in STAN’s User Guide. I will need some time to understand the use of cdf, but fortunately in my simulations I am only using the truncated normal distribution, for which the user guide provides an example!
For the record, while waiting for responses I had also implemented a prior predictive check in R, using the truncdist package for truncated distributions. This prior predictive check seems to have worked well - I could substantially increase the predictive power of one of my models, whose priors on effect sizes were too uniformative and led the model to overfit the data!