P2.1 · ProseMirror ↔ Markdown converter with unit tests

Why this is a separate ticket

Linear stores issue descriptions as ProseMirror JSON, not markdown. The API accepts markdown on write, but reads return ProseMirror. The conversion is the single highest-risk piece of the whole product.

What

Module src/convert/ with two entry points:

• proseMirrorToMarkdown(doc: PMNode): string
• markdownToProseMirror(md: string): PMNode

Nodes to support

• Headings h1–h6
• Paragraphs
• Bold, italic, strikethrough, inline code
• Code blocks with language hint
• Ordered and unordered lists, including nested lists (main gotcha)
• Links
• Linear user mentions (@user → mention node with userId)
• Issue references (KAI-12 → Linear renders as a link)
• Images / attachments
• Blockquotes
• Horizontal rules

Technical approach

• Use prosemirror-markdown as a starting point but override the schema to match Linear's (which differs slightly from commonmark).
• Inspect Linear's schema by pulling a rich issue via the API and logging the returned ProseMirror structure.

Acceptance criteria

• 25+ fixture files: each is a ProseMirror JSON doc. Test: markdownToProseMirror(proseMirrorToMarkdown(fixture)) === fixture structurally (with a deep-equal ignoring whitespace-only differences).
• Lossless roundtrip on 95%+ of fixtures. Document the 5% known-lossy cases in a LIMITATIONS.md.
• Handles nested list depth ≥ 3.