refactor(desktop): register Workbar surface descriptors

[testikun]

Summary

• Register every first-party Workbar tab kind in one exhaustive internal surface descriptor map.
• Route localized labels, icons, launcher rows, and panel rendering through the descriptors while retaining the existing topology registry as the persistence, singleton, shortcut, and ordering authority.
• Preserve lazy panel loading, tab identity and placement behavior, preview/pin/focus/close/reorder actions, and Side Conversation/Terminal resource state.
• Add a structural completeness ratchet plus a real Electron regression for singleton and dynamic tab rendering.

Fixes #4744

Verification

• node --test apps/desktop/dist/main/__tests__/workbar-boundary.test.js apps/desktop/dist/main/__tests__/workbar-model.test.js — 14/14 passed after the base sync.
• npx playwright test --config e2e/playwright.config.ts e2e/session-workbar.spec.ts --grep "registered Workbar descriptors" — 1/1 passed in a real Electron fixture after the base sync.
• npm --workspace @maka/desktop run typecheck — passed after the base sync.
• npm --workspace @maka/desktop run check:architecture -- --base apache/main — 71/71 checker fixtures passed; renderer architecture check passed against apache/main.
• npm --workspace @maka/desktop run build — passed, including renderer entry and third-party notice checks.
• npm run lint and npm run format:check — passed across the repository.
• git diff --check — passed.

Built and launched current head b67c360ce in a visible local Electron App and ran the registered-descriptor journey: the launcher exposed every registered first-party tool, singleton Todo stayed single, and Terminal and Side Conversation opened as distinct dynamic tabs.

Real App screenshot

This direct capture comes from that locally running Electron test; it is not a mockup or generated image.

Base sync

Rebased the single feature commit without conflict onto apache/main at a6fd57ba8, including #4741, which removes the unrelated layout-tier E2E that failed the first CI attempt. No functional Workbar code changed during the rebase.

Review focus

The registry remains private to the Workbar feature. It does not introduce a plugin API or move Side Conversation/Terminal lifecycle ownership into the surface layer.

AI use

Select exactly one:

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

Tool(s) and scope: OpenAI Codex audited the existing Workbar boundaries, implemented the descriptor refactor and regression coverage, rebased the single feature commit onto the updated base, and ran the listed verification. The human contributor remains responsible for review and submission.

Checklist

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

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[testikun]

CI triage: the only failure was e2e/transcript-scroll.spec.ts:712 (turn-fixture-1 → turn-prompt-rail-111). The exact test passed on the original branch both once and in a 10-run repeat (11/11 total). #4707 tracks the identical Linux/Xvfb timing failure and records that an unchanged PR head recovered on a full failed-job rerun. This PR does not touch transcript scrolling or paging, so no unrelated workaround was added.

The single Workbar commit is now rebased without conflict onto apache/main at 19ac204ab, including #4741, which removed that layout-tier E2E from the supported CI baseline. Post-rebase Workbar Node tests (14/14), the focused Electron E2E (1/1), Desktop typecheck/build, and base-aware renderer architecture checks (71/71) all pass. The force-with-lease push triggered a fresh CI run.

CI triage and base-sync update posted with OpenAI Codex assistance on behalf of @testikun.

[testikun]

The post-rebase CI reached Desktop E2E after architecture, lint, format, build, typecheck, knip, and affected workspace tests passed. Its sole failure was the unrelated existing e2e/quote-selection.spec.ts drag-selection case (window.getSelection()?.isCollapsed remained true). On this PR head, the exact test passed 10/10 locally with --repeat-each=10 --workers=1; the Workbar-focused Electron test also passes. No quote-selection or transcript code is touched by this PR. Since main has advanced again, I am rebasing onto the new tip to keep the branch current and trigger a fresh complete run.