Docs: explain native intent styling with executable examples

[noeltock]

Problem

The guide incorrectly described intent trees as unable to carry styling, encouraging unnecessary HTML conversion. It also lacked current alignment and headerless-table examples and treated one image/text mapping as universal. Closes #130.

Solution

Explain explicit native attributes versus interpretation of authored HTML/CSS. Add an executable styling and headerless-table example, plus a short sequential realize snippet using the existing API. Keep generic structural defaults while allowing verified theme layout contracts.

Behaviour / guidance	Read first	Proof
Styled native content stays on the intent route	skills/block-runner/references/GUIDE.md	Corrected routing and native example
Native attributes, layout and headerless cells survive assembly	skills/block-runner/references/ASSEMBLE.md; dev/test/intent.test.ts	Test extracts and executes the documented JSON
Several inputs retain separate finalized reports	docs/reference.md	Executed sequential snippet
Installed skill carries the guidance	skills/block-runner/references	Fresh local installation checked

Diff

+109 / −9 across five files.

 command routing
- styled content -> convert; intent carries no styling
+ explicit native attrs -> assemble
+ existing authored HTML/CSS requiring interpretation -> convert
 assembly guide
+ native colour, spacing, alignment, layout and headerless table example
- image/text rows: never columns
+ Media & Text default; verified theme contracts can require Columns
 library reference
+ sequential realize calls, one report per input

Testing & verification

Reviewed revision: 0590aa5b0d0f33952cd4d8d226e92b86c7576d76 · Environment: Node 24.18.0, macOS arm64.

• npx vitest run dev/test/skill.test.ts dev/test/engine-skill.test.ts dev/test/intent.test.ts: 44 passed, including the documented JSON through realize and parsed-attribute assertions.
• npm run typecheck, git diff --check, and npm run build: passed.
• Extracted and executed the sequential library snippet: two successful reports, one and three valid blocks respectively, zero warnings.
• node dist/cli.js skill --install --dir <fresh-temp-directory>: installed both edited references; commands remain version-pinned.
• Deterministic npm run bench before and after the intent fixes: all 63 SCORE values unchanged.
• Full PR CI passed: https://github.com/humanmade/block-runner/actions/runs/37178345921 (751 tests, four existing skips; Node 20/22/24, package consumers, WordPress proof and scope gate).

Not verified: new model behavior or target-theme/editor presentation. No model benchmark ran; the guide distinguishes pinned headless support from destination proof.

Risk / rollout

Guidance changes the recommended authoring route for explicit native styles. Generic structure and visible-container defaults remain; no runtime feature, batch runner, or expected benchmark tree is added.

Detection: executable guide regression and skill contracts. Rollback: revert this documentation and its example test.