How to read papers without falling into a rabbit hole?

Your problem is quite common among researchers. Actually it's not really a problem, being overwhelmed like this is just natural. Me and all my friends and colleagues face it.

How I overcame this issue: I try to focus on one paper at a time. Try starting reading the latest research paper on a particular subject and go back chronologically. Print the research paper in hard copy, leave your computer and cell phone (if possible) behind and start reading the paper. The point here is to avoid internet access. No matter what question you have, do not search it right away. Write it on the side notes and keep reading. Most of the time the answer will be in the later sections of the same article. Even if you do not get the answers, once you finish reading the paper, you can go online and find all the answers. That's how I read literature without falling in a rabbit hole.

Also, reading the Abstract first, then the conclusion, and then the rest of the paper is also effective.

For me, changing place of study also helps to focus and good instrumental music is always a plus.

(For context, I am in mathematics, and worked on it for almost 20 years before the convenience of the internet...) As suggested by @JonCuster's comment, I think this is what is supposed to happen when one is studying (!). That is, I think genuine study of the literature is "going down the rabbit hole"... and/but not giving up or bailing out somehow. Sure, sometimes there's a "grass is greener on the other side" feeling, but when looking at the research literature it is entirely reasonable to be stubborn in the sense of insisting on explanations.

It struck me that you mention "after an hour, and looking at 2-4 papers"... when I often find myself looking at papers all day long with the conclusion being that there were things that I didn't understand, or wasn't aware of, or had forgotten, that have surprising/significant impacts on the issue I'm considering.

It is true that this (I claim genuine) seeming-inefficiency is wildly incompatible with the assembly-line notion of "research" that we find ourselves being pushed into for various reasons. Hmm... :)

Making notes with good bibliographic indicators, and typing-up these notes rather than leaving them handwritten, and dating everything clearly, is the only thing I know to keep back the deluge. Dating, especially, so that months or years later you can see at least your own personal chronology of awareness is very useful, I think, since it can explain things to your future self about your present and past self's behavior.

In a similar vein, rather than attempting to organize computer files "by concept", in some cases (when the concept is not entirely clear!) it is simply best to organize by chronology.

I have previously written an answer to the math version of the same question. I see no reason why it can't be applied to this case as well.

The human tendency is to prefer worrying about the most recently raised concern: You are reading paper X. You see a term T you don't recognize. You decide to pause reading X to quickly look up T (of course, underestimating how long that will take). You find a paper Y explaining T. While reading Y you find another unclear term U. U has been encountered just now, but T and X were minutes or hours ago - of course U seems like the more urgent thing to figure out. Next thing you know, you're in a meeting about X, you don't know a damn thing about X, but you sure learned everything there is to know about U!

So the human mind, at least in my experience, operates as a stack (in fact a leaky stack, since our attention is quite limited and things at the bottom often become forgotten instead of just delayed). The stack happens to be the defining feature of depth-first search algorithms.

By many relevant metrics, depth-first search (DFS) is a particularly bad choice for reading papers. For instance, presumably the paper you are reading is expected to contain the most important information to you at the time; if there was one that seemed more important, you would read that instead. But with DFS, you will spend a lot of time reading other papers, only distantly related to the original paper - ie. you will waste your time on less useful things.

Because human knowledge is vast, your attention will become exhausted long before DFS hits the wall and starts returning to the original topic.

Probably the biggest reason more efficient approaches (for instance breadth-first search [BFS]) are not more common is that they require additional hardware. Namely, a pen and paper to write a list of things to look up as you read the paper, so that you don't have to split your attention between reading the paper and remembering this growing to-do list. One also has to fight one's own laziness when going to grab a pen and paper, and then (gasp) actually writing, which is much slower than thinking.

Also, the only important point isn't the order in which you look things up. There is also a pruning issue. Many people (including myself) will overestimate how important a term is to understanding the main point (quite a silly habit, since it's essentially trying to guess what it will take to understand a text you have yet neither read nor understood). With a BFS style approach, often it turns out that most of those things you thought you should look up don't really matter and you don't need to look them up. With the DFS, it is much harder to tell what terms actually matter, which ones are irrelevant and which ones become obvious by the end of the text.

Basically, you have to exercise your patience and rebuke your inner sloth. Finish reading your current thing first before looking everything up. Don't worry, you won't forget - just write down what needs to be looked up, with references to where it occurred if you are very worried. Yes, you do have to physically write things, which is clumsy (hence why I say rebuke your inner sloth) but absolutely necessary to overcome some crucial limitations of the human brain. You can type instead of writing, or highlight, or draw !'s on the margins - doesn't really matter, so long as you use a physical (as opposed to mental) means of recording it, and finish what you're reading before starting to read other things.