There's been some good writing on the general failure of "Hypertext books" (e.g. the first couple of paragraphs here [0]). I personally think this stems from some of the reasons that underlie some of those that you outline - books published on the web simultaneously try to be skeuomorphic in retaining the physical book metaphor (organized linearly, mostly static and plain text, etc), while still losing some of the real-world effects of the physical medium (being able to intuitively understand progress, markup, etc). The way that we publish and read text actually hasn't changed much, even though the target medium has of course changed dramatically, and as a result you end up with arguably the worst of both worlds. E-readers try to side step the problem by extending the book metaphor even further, but text on the web doesn't have that privilege.
I'm working on a product, Literal [1], that aims to solve some of your specific problems, specifically providing for a way to annotate and add notes to web content and enabling some degree of source management. My ambition is to move on to solve some of the other problems you raise as well. If you have an Android device and are interested in trying it out, I'd love to connect!
It's unfortunately not a widely supported spec. hypothes.is [0] is the most widely used desktop client, but they don't have a mobile client, and as of last time I checked they don't actually use the data model internally, they just expose a Web Annotation API.
I've heard you loud on clear on the feedback on Literal. I'll have a release out to make sign-in optional in the coming weeks and will follow up here once it's released. I'd be happy to create you some guest credentials in the meantime as well, shoot me an email if you're interested.
I wasn't aware of F-Droid but just created an issue to track publishing releases there here [1]. I'm not sure the inclusion timeline on their side but the work on my side should fast follow guest auth support.
Edit: to clarify _why_ authentication is currently required - the username is used as a primary key on associated data (e.g. annotations), and since I'm actively working on this, I try to personally reach out to everyone via email in order to solicit feedback. You can view the privacy here [2]. I'll move to support an optional auth flow regardless.
Though not textual annotation, some of my favorite recent examples of the value of annotation have been NYT's Close Read series [0], in which they annotate works of visual art as part of communicating an idea or narrative. In some ways it feels like a museum tour or lecture, but more precise (can really call out specific aspects deserving focus, literally _zooming in_ the browser window) and yet also more broad (can easily change subjects, display new visual aids).
I'm currently working on a textual annotation product, Literal [1]. It's open source and implements the W3C Web Annotation spec [2] natively (Yes, there's a web spec for this kind of thing!). You can see a 1m video of how it works here [3]. It's currently Android-only but if you found this article interesting and / or find yourself wishing you could annotate or save bits of text that you read, I'd love to connect with you!