Are unit tests always supposed to run error free, or is that only true for sections of the code that are fairly old and no longer rapidly evolving?

Unit tests should pass for things that the module maintainer considers to be "working". At any point in time, most unit tests should pass, but many may fail in modules that are still be actively developed. In these cases, module maintainers should be marking untested code using the dojo.experimental function. By running the unit tests and watching for "experimental" warnings you should be able to get some sense of which modules are rapidly evolving and which are fairly stable.