[abalone]

Nah, I wouldn't say everyone would flail but Dijkstra. There are simpler solutions to cycle detection.. maybe not as efficient as Floyd, but you could answer the question. For example you could start with something basic and unoptimized like keep a data structure of all the nodes you walk, and when you hit one that's in the structure then you've encountered a cycle. This is a fair approach to engineering solutions to novel problems -- don't panic, get something working, then optimize where necessary.

And if you already know the best practice, like I said, it's not just about whether you know the answer -- it's how simply you can explain it to a teammate and the theory behind why it works. That's also a useful skill to interview for, especially for more experienced candidates.