fix(docx): avoid panic on save, make AnyDocument.save() extension-aware

[pratyush618]

Summary

• DOCX save panic. DocxDocument::save_to_bytes round-tripped through docx-rs's writer, which panics in BuildXML for RunChild on RunChild::InstrTextString (reader-only variant, writer arm is unreachable!()). Any DOCX with PAGE, NUMPAGES, PAGEREF, TOC, or HYPERLINK fields crashed doc.save(...). Upstream issue: bokuweb/docx-rs#750, still open; introduced by PR #528. Fix: return the stashed input bytes — paperjam-docx exposes no DOCX mutation API, so rebuilding was both lossy and panic-prone for no benefit.
• AnyDocument.save(path) ignored the target extension. doc.save("out.pdf") on a DOCX wrote DOCX bytes into a .pdf file. Now routes through convert_to(ext) when the target extension differs from self.format; otherwise writes original bytes unchanged.

Repro (panic + wrong-format file) from the bug report:

import paperjam as pj
doc = pj.open("CurriculumGuideGibbsHighSchool.docx")
doc.save("CurriculumGuideGibbsHighSchool.pdf")  # before: panic; even past the panic, output was DOCX bytes

After this PR: produces a valid 62-page PDF (%PDF-1.7).

Test plan

• uv run maturin develop --release — builds clean
• Re-ran the reported repro on CurriculumGuideGibbsHighSchool.docx — no panic, output opens as a valid PDF with page_count == 62 and extractable text
• Pre-commit (clippy, cargo fmt, ruff, mypy) — passes on both commits
• Existing cargo test --workspace / pytest tests/python/ should remain green (no behavior change for same-format saves; only adds cross-format routing)

Notes / follow-ups

• The DOCX fix should be revisited if/when paperjam grows a DOCX editing API and upstream docx-rs#750 lands — then switching back to inner.build().pack() becomes meaningful.
• The same save(path) UX gap exists on the PDF Document.save() path (writes PDF regardless of extension). Out of scope here; can address in a follow-up if desired.