CmdStan versions and unit testing policy

[amas0]

I ran into failing unit tests during local development and identified that it seemed to be caused by my machine still running cmdstan 2.36 which conflicts with the testing changes introduced in #797. After installing cmdstan 2.37, it's all green.

Do we have a standard approach on how we should handle cases like this? We have a number of tests with explicit checks using cmdstan_version_before, so should we adopt that same approach here? Or should we assume (and state) that unit tests only really need to pass against latest cmdstan versions?

[WardBrian]

In CI we only ever use the latest, and I think it would be too onerous to mark every single test with a bound. The way that I've always described our support for previous versions is 'best-effort', in part because we don't test them

[amas0]

Makes sense - I made a small PR to clarify this for contributors in #831. (Since that is where I looked for this info before I opened this issue).