Yea, I believe this has come up before. The problem is .eex can be embedded into anything: YAML, HTML, JSON, etc. So, in order for the formatter to be able to format .eex files, it would 1) have to know what the surrounding data is and 2) have a formatter for that specific tool in the language.
The first problem could be solvable by some heuristic, but could only solve the problem for the types of files known. It's basically a problem with infinite surface area. The second problem would mean the code base of core would need to include some understanding of every surrounding format. So, because of those reasons, I'd expect something like this to be rejected.
However, if the goal were to make the formatter extensible so that formatting of files other than .ex and .exs can be handled by a third party plugin, that might be more applicable for inclusion into core. That would require some proposed design for how such a thing might work though.