perf(mobile): speed up large Android threads - #5393

Large Android threads spend too much time deriving and mounting feed rows, while a cold thread route swaps header implementations during hydration and visibly moves the title and actions.

This draft is now intentionally mobile-only:

• hide the Android native thread header at route definition time and render loading, unavailable, and hydrated states with the same in-flow header geometry;
• reserve subtitle and action space so hydration changes content without moving the feed;
• enable LegendList item recycling on Android and key recycled row-local feedback to the current content;
• parse feed timestamps once per item before sorting;
• default review highlighting to the JavaScript engine instead of waiting for a failing native probe.

The earlier server history limits, payload preview truncation, projector changes, and associated tests were removed. Sync batching remains owned by #5344; V2 projection/runtime work remains owned by #2829 and #5080.

Verification on the rebased mobile-only diff:

• 17 focused feed and highlighter tests passed;
• targeted lint and formatting passed for all 10 changed files;
• mobile native static analysis passed;
• mobile typecheck reaches two pre-existing errors in untouched SettingsAuthRouteScreen.tsx on current main (onHostBack is missing from the installed auth view prop types).

Earlier physical-Android A/B on the same mobile rendering scope, using a 478-message / 6,696-activity signed-in thread with the 3.01 MB cached detail left intact:

• cold open: branch loaded at ~4.2s; baseline was still loading at 4.28s and next observed loaded at ~11.8s;
• fling frame p99: ~150ms → ~73ms;
• PSS growth: ~161 MB → ~97 MB;
• loading and hydrated title/action bounds stayed fixed.

Those device numbers will be re-run on the rebased head before this draft is marked ready.

Implemented with GPT-5.6 Sol in the T3 Code Codex harness.