Why couldn't you make an HTML5 site with a ten-year-old copy of Dreamweaver? Sure you won't get fancy stuff like code highlighting and suggestions but you could make a HTML5 site with notepad if one really wanted to.
I agree about the clients angle. Clients want to see the images of what the end of the project will look like. Until that changes, mockups will always be with us.
Sure, you could just use it as a text editor but at that point there are better editors you could choose.
Currently I'm loving Jade/Stylus to implement a fairly static website design, and that's just building it with a simple script (with Live.JS to preview changes).
Just to save some time, HTML+CSS+JS fuses, poorly:
1) page description / layout
2) appearance
3) interaction / animation
4) forms
5) content
6) metadata / markup
7) typography
8) vector art and raster images
You could probably throw in HTTP stuff, cookies, how much the DOM sucks, and multimedia too.
Sadly, bad as HTML is, there hasn't been anything better.
I would agree with you that over the years people have pushed features into HTML beyond what it was originally intended for, but that's not what's being discussed.
Contrast to Jade where I would just go: p some text
and let the indentation sort out the DOM structure. It's simple, readable, and doesn't overuse a set of keys placed more or less to be used infrequently (the <> brackets). All this translates to more productivity.
CSS is different problems, but it still stems from being overly verbose (too many vendor tags to worry about) and lacking useful functions, interators etc. Basically CSS sans a preprocessor is just slow and clumsy to work in.
A quick look at HAML and I fail to see how one is more or less verbose than the other. Of course, since I don't use it I suppose I'm not a valid source of comparison.
But to risk negativity, this sounds a great deal like the no semi-colons in Javascript thing a while back.
By the way - no semi-colons in javascript was a good idea aside from a few major factor - it broke things and it was a hack based on loose syntax handling. If it was built into the language design then it would have been a good thing.
What you propose is trivial to the machine, but not necessarily easier for the human.
By the simple fact that no semi-colons broke things and was a hack contradicts the thought that it was a good idea. Building it into the language might be a good idea but it wasn't to try to go around the language.
Also as a designer I find myself telling potential clients that they don't need a designer and should just go with a template or a logo from 99designs. For the longest time I hated doing this, but I realized that if someone wants to spend from $10 to $500 to do something that I wasn't the right person for the gig!
I suspect the reason that this hasn't happen is because HTML has always been an evolving standard. So while you can use your copy of Quark from ten years ago today, you wouldn't be advised to do that with a ten year old copy of Dreamweaver if you want to make an HTML5 site.
Another reason Photoshop comps exist is because of clients who can't comprehend the abstract. As someone who does IA I can tell you that time I've had many clients (who are very intelligent) who just can't go from the abstraction of quickly done wireframes that show navigation to what the final site will be. So until you have clients like that (and maybe they're a generation away) you'll need Photoshop to what that final site will look like. And I say that knowing that Photoshop is really a hack given what the sites can do.