> nobody reads intermediate commit messages one by one on a PR, period.

I do! I find it the easiest way to review code when the author has taken the time to structure it in that way. I'm lucky to work with some great people.