Keep going on warnings in CI and instead mark build broken after it finishes

Motivation

While I find it important that CI checks we keep the code base warning-free, it's very annoying to see a CI build break because I forgot e.g. to remove a variable binding rendered dead by a removal of a trace call.

Proposal

I might be mistaken, but I'd expect GitLab to have functionality for extracting semantic information from job logs. In fact, I think we had means to display the number of broken tests at some point. So surely it must be possible to count the number of/collect warnings and then fail based on that, only after we ran all the tests? This way we get to know which warnings need to be fixed in addition to the tests that broke/possible build errors.