CompatHelper: bump compat for Turing to 0.22, (keep existing compat)

[github-actions[bot]]

This pull request changes the compat entry for the Turing package from 0.20, 0.21 to 0.20, 0.21, 0.22. This keeps the compat entries for earlier versions.

Note: I have not tested your package with this new compat entry. It is your responsibility to make sure that your package tests pass before you merge this pull request.

[storopoli]

@rikhuijzer this is interesting. How does CI fails on 1.6 and works on 1.8. Any thoughts?

[rikhuijzer]

@rikhuijzer this is interesting. How does CI fails on 1.6 and works on 1.8. Any thoughts?

It looks like the values in the chains are different on Julia 1.6. It's probably best to compare the output of Julia 1.6 to 1.8 and step down in both versions to find what exactly is different. Once that root cause is found, it's probably possible to find an issue somewhere at TuringLang. I have to work on something else today, so won't be doing such a dive in the near future

[storopoli]

I'll try to figure out what's the different values and report to TuringLang. Thanks!

[storopoli]

Something is strange in the Poisson model with standardize=true. I am going to merge this for now but I'll debug Julia 1.6 later.