Derive semantic Markdown changes

[hweihwang]

In one sentence: the semantic model that turns two Markdown documents into a typed list of what changed.

What and why

Text cannot tell you what changed between two versions of a Markdown document. Byte diffs report edits, not meaning — nothing says a paragraph was inserted, a table row changed, or a link target moved. This PR adds the model that answers that question: two documents in, a typed list of changes out.

What's in this PR

• A comparison model built on ProseMirror ChangeSet
• Every change classified as text, formatting, attribute, structure, or unknown
• Conservative move detection and position mapping between the two revisions
• An explicit complexity limit — the caller gets a "limited" result, never a silent partial list
• No UI. This unit deliberately has no screenshot; the three views arrive in the next PRs.

Design notes

• One shared schema is an invariant. ProseMirror compares node types by identity, so both editors must parse against the same schema instance — otherwise the result is wrong, not slow. Tests pin this.
• Moves under-report on purpose. Only exact moves are detected; anything ambiguous becomes a remove + insert. A missed move is noise; a wrong move claims edited content is unchanged.
• Attributes and formatting stay separate so "hide formatting-only changes" can never hide a changed link target.

Stack

First of six. #9044 builds on it, then #9045 and #9046. The Collectives side lands in nextcloud/collectives#2697 and #2698 after all four Text PRs.

Verification

ts:check and lint pass; complete suite 1,768/1,768 at this tip. Zero drift against upstream main; the sign-off rebase changed only commit metadata.

AI assistance

Commits carry Assisted-by: Codex:gpt-5.6-sol; an independent review pass was run with Claude Opus 5. This description was drafted with Hermes (deepseek-v4-pro) from my notes, then reviewed by me.

Related

• Show differences (added/removed/changed text) when checking past versions collectives#599
• Version difference extraction #4422