Having spent a lot of time making a similar tool this is pretty difficult. Two issues: (1) highlighting and annotating across (partial) XML with rendered output is very hard without some understanding of the rendering, you can't really just say that a particular highlight is a fixed bounding box or interval; and (2) obviously, web content can change, so even if you could do the above what happens to your annotation highlight from character 5 to 10 when the document shifts?

Personally, I gave up and just converted the web page in question to PDF, and annotated from there. Then you get easy document position markers and no changes over time. I would also be very interested in a system that could do proper webpages natively, though.