Many Elsevier and Springer journals no longer perform copy-editing. Authors are expected to have their paper checked by a native English speaker and copy-edited at their own expense, and then provide the journal with a camera-ready PDF. In my own field, it is the open-access journals published by non-profits that actually have the best language quality and typesetting.
This is a problem that goes beyond journals into for-profit scholarly publishing more generally. In my field, Brill is an infamous publisher for this: it demands camera-ready PDFs for most of the monographs it puts out. So, your library ends up having to spend 400€ on a book where about all the publisher contributed – besides unpaid peer review – is printing, binding, and mailing it out.
I think "needs" might be too strong here. We (the field of CS) get by without it just fine -- the standard practice is to include copy-editing "nits" at the end of one's review. A shepherd, assigned during the final phase, does a final pass on the paper before approving it.
Yup. Typos slip through. There are papers with poor English. Oh well. For the most part, it works out pretty well.
(And as a reviewer, I have little objection to also noting writing fixes while I'm reading your paper. I'm going to spend anywhere from 30 minutes to 5 hours reading the thing -- the writing fixes are a small additional cost. If you've done a decent job on the writing in the first place. If it's totally botched, I'll reject your paper and tell you to fix it before submitting it again. :-)
Now, would I prefer that the authors of submitted papers had to pay $50 for someone to do a copy-editing pass before I reviewed it? Heck yes. But perhaps we'll get DNNs to fix this for us one of these days. :)
Likewise, it's normal for editors in my field to make meta-level suggestions about writing style. I'd wager that history journals place a greater emphasis on prose style than CS journals do, since making an historical argument often depends on telling a compelling narrative. Hence a publishing model that works for a CS journal might not work for humanities journals and vice versa.
Maybe I gave the wrong impression by mentioning apostrophes, however. Professional copy editing is, in my view, super important in terms of differentiating a good journal from a bad or mediocre journal. As is the vetting procedure of finding appropriate peer reviewers and coordination between them. Whether or not you think of that as "real" work compared to producing actual research, it is still work that should be compensated in some way, in my view.
An analogy might be the professor teaching a class vs the maintenance guys who makes sure the projectors are working and the lights turn on. One might be more important than the other, but I don't think either should be working for free.
I also think that you underestimate the importance of copy-editors. It's not the job of a peer reviewer to make sure that the author uses an apostrophe properly, etc. There needs to be a specialist who is dedicated to catching those errors, particularly since, as you note, peer-reviewers aren't directly compensated so it's unfair to burden them with a whole other set of responsibilities.