test: move replay coverage toward stable mobile app patterns

[thymikee]

Summary

The current replay coverage is useful, but too much of the core smoke/behavior validation still depends on Apple system apps and simulator-specific surfaces. That makes the suite less representative of real mobile app development patterns and more sensitive to OS churn and host-environment flakiness.

Proposed Improvements

1. Move most smoke and interaction coverage off system apps and onto a controlled fixture app with stable accessibility identifiers.
2. Keep system-app coverage only where it validates platform-specific behavior:
   • app launch / close
   • deep links
   • permissions / alerts
   • home / app switcher
   • screenshots / recordings
3. Split replay coverage by intent:
   • smoke: fast, minimal confidence checks
   • behavior: selectors, fill, scroll, keyboard, lifecycle, alerts
   • benchmark/soak: longer end-to-end journeys
4. Add flows that better match real mobile app usage:
   • cold start -> deep link -> navigation
   • text entry + keyboard show/dismiss
   • background -> foreground resume
   • modal/sheet open/close
   • permission prompt accept/dismiss/recovery
   • interrupted flow after home/app switcher
   • long-list rediscovery and scroll recovery
5. Add one explicit host-focus regression for iOS simulator so we catch cases where Simulator steals macOS focus during automation.
6. Prefer structural assertions over screenshot dependence in CI where possible.

Why This Matters

• System Settings / Simulator UI changes are outside our control and make otherwise-valid tests flaky.
• A fixture app gives us stable selectors and lets us model the patterns real mobile teams actually care about.
• The iOS simulator focus issue is a host-OS-visible regression and should have dedicated end-to-end coverage.

Suggested Starting Point

1. Build a small fixture/demo app with stable ids and common mobile UI patterns.
2. Migrate smoke and behavior replays from Settings/system-app journeys to that fixture.
3. Keep benchmarks as longer, more realistic user journeys.
4. Add a focused iOS simulator “does not steal host focus” regression test.