There are two problems with this approach. The first one is that it assumes a pure, 100%, contents-only atomic note. The second one is that it creates a couple of extra steps . Expanding on the two problems below.
Atomic notes are not atomic: As mentioned here. Take this very example. Let’s say I have an atomic note that says “Atomic notes are not really atomic, for they have some kind of meta data added to it in a Markdown-based systems like Obsidian”. I would like to add at least two elements to this note: where it came from and other related notes. If I add these bits of information, my so-called atomic note is no longer atomic. It bears not only the content itself, but other information. If I reference this note (by transclusion, for example), I’d also be including its meta data and the links to related notes, which I may not want to do if my interest is only on the core of the note itself.
Extra steps: We are interested in doing exactly the opposite of creating extra steps. Adding extra steps is taxing on ones cognitive system, which degrades understanding and learning. This seems a too expensive side effect. Taking the extra steps argument to the limit, Obsidian itself is not needed. For one can always implement atomic notes in an analog index card system and achieve great results. Luhmann’s intellectual production is a demonstration of that.
In summary, although it is possible to create (cognitive taxing) workarounds and in spite of the technical challenges of developing a block reference system on top of a plain Markdown file structure, it is possible and useful to develop some kind of block reference in Obsidian.