Currently, PDF exports of pages using <score>s ( for example < https://de.wikipedia.org/w/index.php?title=Benutzer:Reiner_Stoppok/Tanzlied_aus_Poniky&oldid=165757690 >from another task) embed them as PNGs.
This makes no sense. If one were to transcribe a PDF sheet music file (e.g. File:Mozart. Zerlina`s aria from Don Giovanni in Ukrainian.pdf ) the result would be worse than the original. The lyrics and other text would not be copyable or searchable, the well known shortcomings of printing raster graphics and so on. Transcribing good quality scans of well preserved sources is not very productive end user wise too (a raster image vs a raster image), even though there are benefits for advanced users.
Basically this makes pages with scores we have badly reusable for readers of whom we cannot demand to be able to extract the Lilypond (or ABC) code and generate PDF externally.
Furthermore generating PDFs is what Lilypond as software does out of box as basically its main functionality. (I do not know about ABC)
I don't know how difficult is it to combine Lilypond's rendering with our PDF export (be it OCG or Electron) but I think it should be doable. At least for the scores used not inline. Small examples of a couple of measures or less which might be used in sentences inline are a smaller concern.
Perhaps one way to improve the situation is related to T49578 but basically it is a separate issue as it is better to use a PDF specific approach.