Sharpen codeblock-ownership.md with what changed since August

[KolesnikovKirill]

What and why

The last unread item, deferred the same way as Playground: recorded and left open rather than implemented. docs/architecture/codeblock-ownership.md already decided this on 2026-08-13 — this strengthens that document, it doesn't replace it (see #17 for why that distinction matters: the first attempt at the Playground write-up made exactly the mistake of not checking for an existing decision first).

What's actually new here

The existing document's reconsideration criterion 2 — "highlighting ownership for browser, Node SSR, and PHP is explicit" — was an open checkbox nobody had answered since August. Two thirds of it are answered now, checked rather than assumed:

Browser and Node SSR: confirmed. VfCodeBlock already calls onServerPrefetch, and Shiki's engine/javascript avoids the WASM Oniguruma engine entirely — highlighting already runs in Node today, no browser required.

PHP: still open, and the reason is worth recording. No native PHP engine reproduces Shiki's output. The one established integration, spatie/shiki-php, doesn't reimplement the engine — it shells out to a real Node process per render. That's the exact "two engines obliged to agree" risk the icon-line decision escaped by precomputing every combination once; it doesn't apply the same way here because code is arbitrary per-request text, not a fixed set of things to precompute.

So the realistic PHP-side choice is between:

• a hard Node.js runtime dependency in Razor (what spatie/shiki-php does), or
• progressive enhancement — Razor renders escaped, working markup (already what "Consumer guidance" in the existing doc prescribes), a controller adds color client-side via Shiki's browser bundle, and a flash of unhighlighted text is the unavoidable cost since there's nothing to precompute.

Progressive enhancement is named as the likely direction, because it's the one consistent with every other adapter in this line. It is not adopted as a decision — no real Razor consumer has asked for this yet, and criteria 1, 3, 4, and 5 in the existing document remain entirely unaddressed.

Also fixed

The check:frozen-showcase acknowledgement removed on the still-unmerged #17 was blocking verify here too — this branch forked from main before that landed. Same one-line fix, applied here as well; no conflict expected when both land.

Checks

npm run verify passes in full.