Writer for `application/pdf`, backed by PDFsharp/MigraDoc. The unified document model is mapped onto a MigraDoc document (sections, paragraphs, tables) and rendered to a real PDF — MigraDoc owns layout, pagination and the low-level PDF object graph, so there is no hand-rolled PDF serialization to maintain. There is intentionally no PDF *reader*: parsing arbitrary PDF back into structured content is out of scope.