Interesting that you mention languages and concept <-> libraries mapping 🤔.

Incidentally, all my job changes involved a change of the entire technological stack :) so I was never really able to map concepts to libraries. Nevertheless, I was able to do it vice-versa - read code in an unfamiliar language, map it to general concepts, and explain solution in this language of concepts.

So what do I think should be asked to validate the candidate is able to understand things with enough detail to be a good hands-on programmer/architect/...?

Rather than providing a topic myself (like mutexes vs semaphores), I let the candidate choose a topic he/she believes to be knowledgeable in and then explore how deep he/she can go. (And I assume the depth and number of "knowledge drill holes" he/she is able to make is a proxy for how well he/she can learn).