See my comment above replying to someone else. There might be some fuzziness, but in context it's easy to classify. And for the weird outliers, the copyright office can make the call on how to classify them. Almost all of these seem pretty straightforward to classify. Just ask yourself what category you would find the work in a library?
>popular scientific books
Certainly nonfiction.
>popular scientific books that embed the science into a nice story
Not sure what you mean, but that sounds like fiction even if it has some science in it.
>hard" science fiction books, which are strongly oriented on physical theories
Same thing.
>Novels that are strongly based on historical facts and even interviews with contemporary witnesses, that nevertheless use some little literary license (for example The Gulag Archipelago
Well your own wikipedia link classifies it as fiction at the bottom. In fact all the examples of novels that are loosely based on truth are fiction.
But no one is ever going to claim a scientific paper is a work of fiction, and should be protected for 100+ years.
>A scientific description of some ethical thought experiment
This seems like the only fuzzy example. But such a thought experiment will almost always be in the context of a larger nonfiction work. And it will be published in a place like a scientific journal with other nonfiction, and be placed in the nonfiction section of a library. So that seems like the most practical and natural classification, legally.