fix: real insert/remove_watermark, honest image-watermark error, sectPr hydration

[ebolamerican]

Summary

The watermark tools were stubs that reported success without touching the document, and open_document never populated sectionProperties from the source sectPr — so any typed re-emission of word/document.xml collapsed margins, page geometry, and header/footer references to defaults.

This PR makes those tools do what they claim, and works around the DocxReader sectPr gap server-side.

Changes

• insert_watermark actually inserts a watermark. Writes Word's own VML WordArt watermark (PowerPlusWaterMarkObject textpath shape, o:spt="136") into the default page header, honoring font / color / size / semitransparent / rotation. Three paths:
  • Existing default header — append the watermark paragraph to the header's typed bodyChildren.
  • Scratch-built document (no archiveTempDir) — pure typed path via addHeader; the writer emits every part from the typed model.
  • Source-loaded document with no header part (the reported repro: an engine-built .docx with footer1.xml only) — wire the new header surgically into the preserved archive (relationship in word/_rels/document.xml.rels, Override in [Content_Types].xml, <w:headerReference> inserted into the existing body-level <w:sectPr>) before registering the typed Header, so DocxWriter's overlay gates stay closed and the source rels / [Content_Types].xml are preserved verbatim. A second insert errors instead of stacking duplicate shapes.
• remove_watermark actually removes. Strips watermark paragraphs from every header (typed raw blocks and source-parsed w:pict runs) and errors honestly when no watermark exists.
• insert_image_watermark returns an honest error. Image watermarks need media-part plumbing that isn't implemented; the tool now returns a not-implemented error instead of the prior false success that wrote nothing.
• open_document hydrates sectionProperties (header/footer refs incl. first/even, page size + orientation, margins, columns, titlePg, section break type, pgNumType, vAlign, line numbers, docGrid) from the source body-level <w:sectPr>. This works around DocxReader leaving sectionProperties at defaults (see filed upstream issue), which otherwise caused every typed re-emission of word/document.xml to silently drop the source's footer reference, custom margins, and page geometry.

Tests

• Tests/CheWordMCPTests/WatermarkInsertRegressionTests.swift — append-to-existing-header, create-header-when-none-exists, remove actually removes, second insert errors.
• Tests/CheWordMCPTests/SectPrPreservationProbeTests.swift — insert_paragraph preserves the source footer reference.

swift build clean; swift test green (5/5 on the new suites).

Related upstream issues

These fixes work around, but do not fix, three root-cause bugs filed separately:

• DocxWriter.buildTypedRelationships hardcodes rId1-3; drops/remaps source styles.xml / numbering.xml rels on overlay re-emit ooxml-swift#83 — DocxWriter.buildTypedRelationships hardcodes rId1–3 and drops/remaps source styles.xml / numbering.xml relationships.
• DocxReader never populates WordDocument.sectionProperties from the body-level <w:sectPr> ooxml-swift#84 — DocxReader never populates sectionProperties from <w:sectPr>.
• protect_document is a stub — reports success without writing <w:documentProtection> #172 — protect_document is still a stub that reports success without writing <w:documentProtection>.

🤖 Generated with Claude Code