I'd go for the SVG format instead of rasterizing to PNG. It's going to be lighter and give you a similar crisp feeling.

Have you tried passing the pages through a OCR transcription system like tesseract? You also get this with the text-parsing feature of the reMarkable 2 for free, and I was thinking you could use the [link title] syntax to mark links and have some sort of automatic mapping to the links so you don't have to specify the links "by hand."

Anyhow, great idea. It's a bit small to read on iPhone.

A cool idea would be to convert each word to an individual image and put each of them inside of an inline div to make the article responsive.