Because the PDFs only carry an image? (I'm assuming if there was real text the tools could extract it, but could be wrong)

PDF to text tools can extract some text, but get confused by tables, maths, etc.

You can annotate a PDF with the raw actual text, which is then used by screen readers, but typst (and LaTeX currently) don't do this. It certainly isn't trivial to do, but it would be useful.

The other option would of course be to produce HTML, which (unless you do horrible things) is generally very accessible, but neither typst or LaTeX do this well (LaTeX does it OK, but I find I usually need to clean up the result).

PDFs produced by LaTeX aren't accessible either.

You are right, but LaTeX has the advantage of having a much bigger user base and documentation, so I'll still with it until Typst has some clear advantages (like accessibility).