feat(cli): print the render's reference_background as an @hf-result line - #7203

What

A piped, non-quiet local render now prints one @hf-result JSON line after it succeeds, carrying the reference_background the render already measured. Batch manifest rows carry the same reference_background value.

@hf-result {"version":1,"operation":"render","reference_background":{"kind":"solid","rgb":[12,34,56]}}

reference_background is exactly one of {"kind":"solid","rgb":[r,g,b]}, {"kind":"none"}, {"kind":"unsupported_non_solid"}, {"kind":"unknown"}: the object renderLocal / renderRange already return, re-projected through referenceBackgroundResult at the writer. No new computation.

Why

renderLocal returns reference_background, but nothing printed it: the single-render path discarded the returned object and batch rows copied only durationMs and renderTimeMs. A caller that runs hyperframes render as a subprocess (piped stdout, no --json) could not read the measured background. This is the same transport as @hf-diagnostic and @hf-progress, so it needs no new flag.

Related work

Builds on the measurement merged in #7007. Refs: Cut Room harness row D56-24 (blank scan should compare against the composition's own background).

How

• packages/cli/src/ui/renderResult.ts: renderMachineResult, written like renderMachineDiagnostic: silent on a terminal or when HYPERFRAMES_STDOUT_IS_TTY=1, silent in quiet mode (including batch JSON), one line, version 1.
• executeHeldRender prints it once from the value the single-render executor returns. A failed render throws before the call, so a failure never prints it. A --docker launcher gets no field back and prints nothing.
• BatchManifestRow.reference_background is null until the row completes and stays null for a failed row. Manifest version stays 1 (additive key).
• packages/cli/README.md documents the line next to @hf-diagnostic.

Changeset / release note: none needed for the PR; release notes are drafted by release:prepare.

Test plan

• Unit tests added/updated
• Manual testing performed: not done beyond tests (no real browser render, see below)
• Documentation updated (if applicable)
• Comments follow CONTRIBUTING.md "Comments"

Tests that fail on the old code and pass on this head (checked by restoring only execute.ts and batchRender.ts to origin/main):

• execute.test.ts > "the machine result line after a single render" > prints the render's reference background once (solid, none, unsupported_non_solid, unknown): expected one line, got none.
• batchRender.test.ts > "copies each row's reference background into the manifest and leaves a failed row null".

Also covered, passing on both: no line on a failed render, no line in quiet mode, no line when a Docker render reports no background (execute.test.ts); line shape, single line, malformed value re-projected to unknown, silent on terminal / HYPERFRAMES_STDOUT_IS_TTY=1 / quiet (renderResult.test.ts, new module).

Ran locally: renderResult.test.ts, execute.test.ts, batchRender.test.ts (38 tests pass), oxfmt, oxlint, tsc --noEmit on the cli package (only pre-existing @hyperframes/inspector/live errors from an unbuilt package), check-comment-citations.mjs.

Could not verify

• No real browser render was run, so the line was not observed from a live hyperframes render process; the single-render test drives executeRenderPlan with a stub executor that returns each of the four kinds.
• comment-ratchet.mjs needs full history; CI runs it.

Authored by Jerrai.