ui-react: pair the remaining Cartesian theme cells in answer-delivery and journal runtime tests #3329

Context

#2763's acceptance audit of #1849 found two runtime families that still run the full 320/1280 × dark/light Cartesian matrix. The same files, and docs/process/theme-parity.md, reduce other families to the paired cells 320 dark and 1280 light. #1849's definition of done says "Runtime suites preserve independent failure-mode coverage without an automatic Cartesian theme multiplier", and "Remaining genuine theme-specific checks state a concrete reason". Neither family states a theme-specific reason, and both were added after the theme reduction:

• packages/ui-react/tests/answer-delivery-runtime.test.ts:337-340: for (const theme of ["dark", "light"]) × for (const width of [320, 1280]) around 1887: Other has explicit current-text Submit at ${width}px in ${theme}. It came in through CONTRACT: fix(ui-react): keep explicit Submit for Other answers #1896 (2026-10-15). The only theme-dependent assertion is OTHER_LAYOUT_USES_REAL_THEME (computed color-scheme). Submit count/disable, keyboard reach, the ≥44px touch height, no overflow, the tap/Space path and single settlement do not vary by theme in the source.
• packages/ui-react/tests/turn-journal-runtime.test.ts:101: for (const width of [320, 1280]) for (const theme of ["dark", "light"]) around journal configuration coverage keeps small changes first and preserves inspection. It came in through CONTRACT: keep retained journal changes ahead of coverage gaps #2214 (2026-10-16). The same file already defines journalRuntimeCells = [[320, "dark"], [1280, "light"]] (line 100), and the comment at lines 97-99 says JournalTheme stories cover both themes.

Hosted cost at exhaustive scheduled run 41382713082 (source 8e9855e706ab43dc1bb5b822296f171431e2546d): eight cases, 22.33 s summed case time. Removing the two off-diagonal copies of each family saves about 11 s. The cost is small; the finding is a definition-of-done gap.

Scope

For each of the two families, choose one, with an assertion map in the PR:

1. Run only the paired cells (320 dark, 1280 light), as the other families in these files do. Name the central Storybook/consumer story assertions (docs/process/theme-parity.md, packages/ui-kit/stories/conversation/JournalTheme.stories.tsx and the answer/Other consumer stories, if any) that cover the removed copies' theme-only assertions.
2. Keep the Cartesian copies, with a source comment and a docs/process/theme-parity.md entry that state the concrete theme-specific failure mode each removed copy would miss.

If no central story covers the Other answer card's theme-only layout in both themes, add the minimal consumer story assertion rather than keeping a runtime copy. Otherwise state why option 2 applies.

Out of scope

• Theme loops in browser/layout/pointer/visual projects (packages/ui-react/tests/browser/**, packages/ui-kit/tests/visual/**). Whether those are in Epic: tests give faster feedback with less redundant work #1849's scope is a maintainer question recorded on Epic: tests give faster feedback with less redundant work #1849.
• Accessibility and contrast cases that are inherently theme-specific (tests/prose-links-render.test.tsx, tests/answer-suggestions-targets.test.tsx, packages/ui-kit/tests/contrast.test.ts).
• Any new lint/guard against theme multipliers.
• Any change to the delivered assertions themselves beyond removing redundant copies.

Acceptance criteria

• Each family runs the paired cells, or carries a stated concrete theme-specific reason in source and in docs/process/theme-parity.md.
• For a reduction, the PR maps every assertion of the removed copies to a retained runtime cell or a named central story assertion. A mutation is recorded: break a theme-only style in the covered component, show the central story assertion fail, restore.
• All other unique runtime assertions (settlement, frames, Revert/Undo effects, download bytes, keyboard and touch paths) still run at both widths.

Verification

bun run test packages/ui-react/tests/answer-delivery-runtime.test.ts packages/ui-react/tests/turn-journal-runtime.test.ts and the affected Storybook projects (node scripts/visual.mjs --project=storybook --project=storybook-light) under the pinned offline harness, then hosted ready-candidate proof. Record before/after case counts.

Contract impact

None. Test structure only.

Blocked by #3291