Strengthen the playground-ownership decision with technical evidence

[KolesnikovKirill]

What and why

Playground was the last unread item on the "not yet ported" list. This closes the question with evidence, not a new decision — docs/architecture/playground-ownership.md already decided it on 2026-08-13. This branch strengthens that document rather than duplicating it (see below).

Why this isn't "the icon question again"

Every other retained piece turned out to be portable once read closely enough. Icons looked like a 2,135-line rendering engine nobody wanted to duplicate, and turned out to be static data once precomputed — 116 icons, a fixed number of combinations, computed once and shipped to both platforms.

Playground doesn't have that shape:

self.addEventListener('message', (event) => {
  const request = event.data;
  // ... transpiles request.sources with the TypeScript compiler ...
  self.postMessage({ type: 'result', id: request.id, outputs: [] });
});

export function runInIframe(iframe: HTMLIFrameElement, html: string): void {
  iframe.setAttribute('sandbox', 'allow-scripts');
  iframe.srcdoc = html;
}

Worker and HTMLIFrameElement aren't a framework choice Vue happened to make — they're what a live, reactive, sandboxed preview of arbitrary just-typed code requires. A PHP request answers one request with one response; it can't host a running session watching for the next keystroke. An icon's output is a pure function of three known values. Playground's output is a pure function of whatever someone is typing right now — not a harder version of the same precomputation problem, a different one.

Nothing in the CodeMonster line (ui-vue, ui-layouts, ui-runtime, ui-icons) depends on any of the three playground packages — confirmed by grep, not assumed.

A correction made mid-branch

The first version of this PR added a new file, playground-line-ownership.md, without checking whether this question had already been decided. It had — playground-ownership.md, 2026-08-13, with a thorough package-boundary and threat-model argument reaching the same conclusion. A second file saying the same thing under a near-identical name would have fragmented one decision across two documents that don't reference each other, which is exactly the kind of stale-pointer problem this repo has spent the past several branches correcting.

The fix: the new technical evidence (the two snippets above, and the explicit contrast with icons) is now a section inside the existing document, under a Revisited: 2026-09-03 line, rather than a rewrite of reasoning that was already correct.

What's in the PR now

• docs/architecture/playground-ownership.md gains a "Why this held up under the same test that moved icons and layouts" section.
• The three retain entries in the maturity backlog get reasons that cite that document.
• CodeBlock is deliberately left out — it has no comparable browser-only primitive forcing the same conclusion (Shiki already runs server-side in Node during Vue SSR) and is a real, open engineering tradeoff rather than a settled question. It's getting its own read in a follow-up.

Checks

npm run verify passes in full.