Thank you very much for your (very fast) answer! Unfortunately I have only now had time to play around with the code.
It is not exactly what I wanted but it helped me anyway. You were right that it was helpful to know how to DIY the plots.
I work with some data (laboratory results) for which there are relationships in the literature. The relationships are based on a linear regression where the independent variable is logarithmically transformed (such as here: https://kenbenoit.net/assets/courses/me104/logmodels2.pdf).
Therefore I assume for my observed y: Y_i = α + βlog(X_i) + ε
So the linear scaled plot shows a nonlinear relationship and in the log-linear scaled plot it gives a linear relationship.
I adapted your code for this model and it worked for me (see graphic below). But I think I had to add var_names = [“mu”, “y”] for sampling the posterior_predictive, otherwise the code for reg_line_samples did not work.
Additionally I added in analogy hdi_mu = az.hdi(idata_spaced.predictions).mu for the credibility interval of the mean and it matched the regression lines of the sample.
When I tried to get this right graphic with the az.plot_hdi-method, I got this strange result displayed in the second graphic in my first question on the left side.
But I think I will use the method you suggested in future. Thank you very much!