fix(desktop): keep WorkHub chat bubbles on --radius-chat

[liuxiaocs7]

Summary

WorkHub's conversation surface passed density="compact" into Astryx's ChatMessage / ChatMessageList, and ChatMessageBubble maps that to border-radius: var(--radius-container) (12px). The WorkHub composer stayed on --radius-chat (28px), so a bubble and the dock on the same surface disagreed by more than 2x. The main transcript already avoids this — #3452 dropped the compact density from its ChatMessageList; WorkHub reintroduced it in #3497.

This drops density="compact" from WorkHub's chat primitives — both ChatMessageList calls and the user/assistant ChatMessage rows in WorkHubMessageFrame — so the bubble falls back to the primitive's --radius-chat and rounds together with the composer as one conversation surface, exactly as #3452 fixed the main transcript. gap={4} stays: it sets the row gap explicitly and never depended on density. No product CSS override is added; chat-message.css already documents that the bubble radius is left to the primitive on purpose.

Fixes #4914

Verification

Before / after on the real story render (Product/WorkHub → SubmittedWorkKeepsTargetMetadataInside), light and dark:

Computed border-top-left-radius, measured in the running Storybook (Playwright):

State	User bubble	Composer plate
Before (density="compact")	12px	28px
After (this PR)	28px	28px

Light and dark measured identical. After the fix the user bubble and the composer plate agree at 28px — matching the main transcript.

• Regression coverage (Storybook play): the existing SubmittedWorkKeepsTargetMetadataInside play now asserts the projected user bubble's computed border-top-left-radius equals the real composer plate's — compared element-to-element, so the bubble and dock on one surface must round together or it fails. This lives in the story play per e2e-budget.json (layout geometry / pure CSS is not an Electron-tier concern), the same tier test(desktop): finish moving non-Electron coverage out of the E2E tier #4877 moved the sibling Side Chat / WorkHub geometry into. Red without the fix (bubble 12px vs plate 28px); I ran the story both ways to confirm.
• Biome (biome check on the changed files): pass, no fixes.
• Typecheck (tsc -p tsconfig.renderer.json and tsconfig.storybook.json): no errors in workhub-surface.tsx or workhub.stories.tsx. Remaining diagnostics in the run are pre-existing/unrelated, because this worktree had no installed deps — I ran against a lockfile-identical sibling node_modules.
• Renderer architecture snapshot: unchanged — the edit removes a prop only; all imports/hooks/deps recorded for workhub-surface.tsx are untouched, so no renderer-architecture.json regeneration is needed. No .tsx/.css added or renamed, so the surface-inventory gate is not triggered.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Code — diagnosed the issue, made the fix, added the story-play regression assertion, and captured the before/after screenshots. Commits carry a Generated-by: Claude Code trailer.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Biome (lint/format) and typecheck of the changed files pass locally; the SubmittedWorkKeepsTargetMetadataInside play was rendered both ways (fixed = green, density="compact" restored = red) via Playwright against the running Storybook.

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Reviewed current head 531249fd (OPEN, MERGEABLE, awaiting human review). Technical GO — no P0–P3. CI test fully green on this head (12m39s, including Electron e2e budget check, Test Storybook audit contracts, Build Storybook, and Storybook smoke).

What changed

Four WorkHub chat primitives (two ChatMessageLists, user/assistant ChatMessages) drop density="compact", so bubbles fall back from --radius-container (12px) to the primitive default --radius-chat (28px), matching the composer panel — the same fix previously applied to the main transcript. gap={4} is correctly kept (row gap never depended on density). No CSS override was added (chat-message.css:100 deliberately leaves radius to the primitive; an override would conflict).

Correctness points verified

• The density blast radius was checked at its widest point: density=compact also drives Markdown heading scale and list/quote rhythm, but WorkHub message rows render plain <p> text with no Markdown body — on this surface density only ever affected bubble radius, so removing it has no collateral density regression.
• Before/after screenshots (light and dark) confirm the claim: 12px vs 28px mismatch before, identical bubbles after, with no other movement in metadata blocks, spacing, or text.
• The extended story asserts border-top-left-radius equality against the real composer plate element rather than a 28px literal — both move together (or the test fails) if the upstream token changes, which is the right way to pin it. The e2e-budget annotation correctly attributes pure CSS geometry to the Storybook layer.
• Six-layer pass: design (real issue, 12 vs 28 on one surface) → function (real WorkHub submission path) → complexity (removing a knob instead of adding an override, entropy down) → tests → naming/comments → style.

Verification boundary

• No real WorkHub submission was driven end to end; geometry is verified via screenshots plus the element-to-element regression, not a live session.

---

Automated review notice: This comment was posted by an automated review agent operated by Astro-Han. It is not an independent human review and does not replace one.

简体中文

评审结论来自自动化审查流程；发布者没有读这份 diff，核的是当前 head 有没有漂移、以及 exact-head 的门禁状态。当前 head 是 531249f，未关闭，检查全绿。技术上无阻断问题。合并由人类定。