docs: add SUPPORT.md and clarify known conversion limits

[HaoChiBao]

Summary

• Add SUPPORT.md documenting what each input format keeps, reduces, or drops (images as alt text, headers/footers, spreadsheet visibility and number formats, nested tables, PDF model/OCR limits, and more), grounded in current parser behavior.
• Link that page from the README Supported formats section and from the convert-documents agent skill.
• Document that literal line-start 1. in plain paragraphs is escaped as 1\. on purpose so Markdown does not invent a list. Real Word w:numPr lists still render as normal ordered lists. Closes the confusion in docx to md puts a backslash between number and period in ordered lists #41 after inspecting the attached sample (no numbering part / no w:numPr).
• Note the pdf-inspector = 0.1.8 floor in Cargo.toml so the RUSTSEC-2026-0187 lopdf fix (0.42.0) is not lowered by accident. main already resolves the fixed versions; transitive DoS via pdf-inspector -> lopdf 0.41.0 (RUSTSEC-2026-0187) #67 can close with this clarification.

Closes #2
Closes #41
Closes #67

Test plan

• Skim SUPPORT.md against README Features / Supported formats for tone and accuracy
• Spot-check a few claims against src/formats/** (headers/footers, PDF to_document, spreadsheet number formats)
• Confirm README and skill links to SUPPORT.md resolve
• Confirm Cargo.toml still requires pdf-inspector = 0.1.8 and cargo tree -i lopdf --locked shows only 0.42.0
• Optional: convert the docx to md puts a backslash between number and period in ordered lists #41 sample and confirm 1\. output remains expected for literal paragraph text

---

Summary by cubic

Documents per-format conversion limits in SUPPORT.md and links it from the README and the convert-documents skill. Clarifies the intentional escaping of literal line-start list markers and records the pdf-inspector 0.1.8 floor to keep the lopdf RUSTSEC fix. No behavior change; this documents current behavior.

• Verify SUPPORT.md matches current parsers (images as alt text, headers/footers omitted, nested table flattening, spreadsheet number formats/visibility, PDF to_document unsupported and no OCR).
• Check links from README.md and skills/convert-documents-to-markdown/SKILL.md resolve to SUPPORT.md.
• Confirm Cargo.toml retains pdf-inspector = "0.1.8" so lopdf >= 0.42.0 stays resolved.
• No migration or API changes.

^{Written for commit f4e2e4c. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 4 files

_{Re-trigger cubic}