Add .rtf (Rich Text Format) read+write support

[Mearman]

Add read+write support for .rtf, Microsoft's Rich Text Format. RTF is the cleanest pivot fit of any format documents.js doesn't yet support: it's structurally wordprocessing (paragraphs, runs, character properties, paragraph properties, tables, lists, images all have direct ContentDocument equivalents), and it can express more of the wordprocessing variant than markdown can (colour, font family/size, alignment) — no document-schema.js model changes needed.

The codec itself is a real, multi-week hand-written undertaking: RTF is tokenised plain text, not XML, so none of the existing XML-parser plumbing (ooxml.js, odf.js) is reusable. It needs its own destination/group state machine, \uN/\ucN unicode handling with codepage fallback, and parsers for five header mini-formats (fonttbl/colortbl/stylesheet/listtable/listoverridetable).

Previously closed as not-planned for niche demand under the hand-write-no-third-party-library stance; reopened as full support is now wanted.

[Mearman]

Closing as a recorded decision (not planned / out of scope) per the assessment in the body. Re-open if a concrete need changes the verdict.

[Mearman]

Opened #843 (draft) with packages/rtf-codec — a hand-written reader and writer against the RTF 1.9.1 specification. Leaving this issue open, because the PR does not cover everything asked for here.

What shipped. The tokenizer and destination/group state machine the issue anticipated, built literally on the spec's own "Conventions of an RTF Reader" model, plus \uN/\ucN with the ANSI-fallback skip (including the "control word counts as one character" and "a brace ends the run early" rules), \'hh, \binN, and code-page decoding across \ansi/\mac/\pc/\pca, \ansicpgN and a font's own \cpgN/\fcharsetN. All five header mini-formats are parsed: \fonttbl, \colortbl, \stylesheet, \listtable and \listoverridetable, with \lsN resolved through the override table to the list's own level formats. Body coverage is paragraphs, runs, tables, lists, PNG/JPEG pictures, HYPERLINK fields and the special-character vocabulary — read and write both, with the writer minting the header tables from what the document actually uses and emitting deterministic pure-ASCII output.

What is still open.

• Registry wiring. rtf is not yet a DocumentFormat member, so it is not reachable from convertDocument, the CLI, the MCP server or the web UI. That touches five tables in documents.js and three exhaustive conversion-matrix tests, and is its own change — epub-codec did the same (Wire epub into the conversion engine, CLI, MCP server, and web UI #802).
• listoverridetable override levels. The \lfolevel/\listoverrideformatN/\listoverridestartat semantics the issue flagged as a likely corner are not implemented: the override is followed to its \listidN and the list's own levels are used, so a start-at or format override on a particular level is ignored.
• Fidelity constructs. Bookmarks, tracked changes and content controls all have RTF spellings and are not mapped onto the harmonised constructStart/constructEnd vocabulary in either direction.
• The residue channel. document-schema.js's SourceFormatSchema has no rtf member, so the codec has no legal residue value to emit at all. That is an additive one-line change to that package, and probably wants doing before any restorable-fidelity work here.
• Notes and page furniture. Footnotes, endnotes, headers, footers and annotations are read past with a diagnostic — the flat ContentDocument has no position for them, and a footnote's real home is the tree-only definitions table a codec producing the flat form cannot reach.
• Format and construct gaps listed in full in the package README's own table: East Asian DBCS code pages, code page 42, metafile/bitmap pictures, nested tables, embedded and drawing objects, multiple sections, superscript/subscript, cell borders and vertical merge, right-to-left text.

Every one of those drops reports a named diagnostic rather than disappearing, and a coverage suite fails if any code in the table has no reachable emit site.

[Mearman]

rtf-codec is real read+write now: destination/group state machine, \uN/\ucN unicode with codepage fallback, and the five header mini-formats (fonttbl/colortbl/stylesheet/listtable/listoverridetable), wired into documents.js's composition engine (docx/odt/markdown bridges, PDF pivot via a docx bridge, construct mapping for bookmarks/revisions/anchors, cell borders/shading/merges). Scope and every deliberately-unhandled construct are documented in packages/rtf-codec/README.md.