I think these are very good questions. It’s often hard to know when to stop in knowledge management. When do I know if my notes are too sophisticated to be worth it?
My approach to answer this type of questions is based on my objectives. I feel like the most important tool to say “I should stop here” is to know what I take notes for.
I’m a knowledge worker. In my case, the reason why I make notes is to resurface information when I need it. The gap between storage and retrieval will vary from note to note, and goes from minutes to years. In the minute range, I have the link to that table I’m working out today. In the year range I have my higih-level notes on “algorithm X” and its caveats, in case I need them for a new problem moving forward.
The notes I take for a few hours are not very elaborate, and I mostly make them directly in my daily notes. The notes that I store for years need much more work and context written in them, so they stay usable for longer.
I hope I summarized my approach well enough. Now I’ll try and use it to answer your questions:
Is this one atomic idea, multiple ideas, or none?
I would ask: am I likely going to want to retrieve these informations together, or do I want to be able to retrieve them separately? If I likely want to retrieve them together, they are probably atomic.
Is this too obvious to even count as an idea?
When you want to retrieve it, are you very likely to know it by heart, or do you think there’s a chance you might need your note to remember it fully?
How do I know if this should be a backlink or an outgoing link?
I think this is a matter of preference. The reason I like backlinks is related to the analysis paralysis you express: it makes it less important where you store information. If I want to write something down but I’m not sure where, I just put it in my daily note. I make sure I link to all the relevant topics such as “Transformer Architecture” or “Gradient Descent”. When I want to resurface it, I look at the corresponding backlinks.
Am I creating too many links?
Are the links making information retrieval difficult? Digital tools like Obsidian are very resilient to a lot of annotation, because you can always filter it through search later. I tend to link more than less.
Do my links even make sense?
Only you can tell 
Do they help you surface information when/if you need it?
Why isn’t staring at the graph telling me anything useful?
My graph doesn’t mean much either. At best, its clusters tell me what areas of notes are heavily interconnected in my vault. Other people have better use of the graph, but they tend to be very deliberate about the links they place. I tend to link a lot and not worry about the graph too much. Also, my graph view is a lot nicer when I remove some parts of my vault, like my daily notes.
In conclusion, I advise you not to worry too much about it. Information management is hard, and it is why we need software like obsidian. Do what feels reasonable, you can always edit text files later.
Cheers
The graph is mostly useful for making pretty pictures. Some people find it useful for other things but many don’t. I don’t use it myself. I hear the filters can help and local graph view is more useful.