perf(test): remove redundant desktop verification setup

[Astro-Han]

Part of #2382.

Problem

Desktop verification repeatedly constructs the same expensive state:

• Multiple Playwright tests launch Electron with identical long-transcript or long-sidebar fixtures.
• Storybook smoke repeats viewport and color combinations that add no distinct assertion.

Increasing global worker concurrency would not remove this duplicated work and could reintroduce focus races.

Outcome

• Merge compatible assertions using the same Electron fixture into coherent journeys.
• Preserve distinct IPC, focus, scrolling, remount, persistence, lifecycle, and animation behavior.
• Move final-state layout and text matrices to the cheapest suitable contract.
• Keep the minimum Storybook matrix required by declared checks and retain catalog mounting coverage.
• Keep Electron E2E and Storybook changes in separate PRs.

Include before/after test counts, page or Electron startup counts, and comparable wall time.

[UncertaintyDeterminesYou4ndMe]

I'll take this one, continuing the #2382 series (#2387 is in review as #2476/#2480).

Plan per the issue outcome list:

1. Inventory every Electron launch across apps/desktop/e2e — which fixtures are seeded, which tests share an identical fixture+launch, and what each test asserts — plus the Storybook smoke matrix and its declared checks.
2. Merge compatible assertions that relaunch the same seeded state into coherent journeys, preserving distinct IPC, focus, scrolling, remount, persistence, lifecycle, and animation contracts as separate launches where the contract demands a fresh process.
3. Move final-state layout and text matrices to the cheapest contract that still proves them.
4. Separate PRs for Electron E2E and Storybook, each with before/after test counts, launch counts, and wall time.

[UncertaintyDeterminesYou4ndMe]

Both PRs are up:

• test(storybook): shrink the smoke matrix to its declared checks #2481 — Storybook: manifest matrix 92 → 48 serial jobs via the validator's own viewportOptOuts mechanism (every drop carries a written reason); catalog mount sweep untouched at 100 renders. Local smoke 43.0s → 28.6s (−33%).
• test(desktop): merge same-fixture e2e assertions into coherent journeys #2486 — Electron E2E: 69 tests / 69 launches → 34 / 34 by merging same-fixture assertions into journeys with documented ordering constraints; 11 orphan window fixtures deleted. workers=1 (CI-like): 192.4s → 147.2s (−23%); workers=4 local: 57.9s → ~43s. All merges annotated; four deliberate non-merges explained in the PR (one backed by an observed shared-window failure mode).

Remaining candidates deliberately left as maintainer decisions rather than claimed here: the cross-file invocableSkillsWindow family reorganization (5 files, 12 tests — would move regression ownership between files) and the sandboxBoundaryWindow pair.