test(mobile): cover the AbortSignal polyfill (CODE-478)

[AprilNEA]

Summary

apps/mobile/src/polyfills.ts (CODE-462) is what lets the mobile client open a socket at all, and it had no test. This adds apps/mobile/src/__tests__/polyfills.test.ts — four cases in the app's existing vitest project, no production code touched.

The module patches at import time behind typeof AbortSignal.prototype.throwIfAborted !== "function", with no exported installer, so each case hands it a runtime without the API (delete throwIfAborted + reason off the prototype) and re-runs it through vi.resetModules() + dynamic import(). afterEach puts all three patched globals back, AbortController.prototype.abort included.

Cases:

1. throwIfAborted is installed when the runtime lacks it, and a live signal does not throw.
2. The reason round-trip — abort(reason) → signal.reason → throwIfAborted() rethrows the same value by identity, with a non-Error reason so a future narrowing would fail here. RN 0.86 discards the argument to abort(reason) entirely, and ConnectionController re-throws signal.reason when a run is superseded, so the symbol-recording wrapper is the only thing keeping that value alive on device.
3. abort() with no argument yields a foxts AbortError, from both signal.reason and the throw.
4. A runtime that already has the Abort API is left untouched — descriptor identity for throwIfAborted, reason, and abort. This is the guard that will let CODE-477 delete the module on RN 0.87 / Expo SDK 58 without ceremony.

Salvaged from the closed PR #309 (CODE-466, a duplicate of CODE-462): that branch shipped a second, weaker polyfill; only its test was worth keeping, and it is rewritten here against the surviving module.

Verification

• pnpm test polyfills — 4 passed.
• pnpm test — 256 files passed, 3 skipped; 2,082 tests passed, 5 skipped.
• pnpm format:check, pnpm lint (0 errors), pnpm typecheck — all pass.
• Mutation-checked, since a test for a polyfill is easy to write vacuously. Removing the AbortController.prototype.abort wrapper fails cases 2 and 3 only; forcing the guard always-true fails case 4 only. Both mutations were reverted.

No device run: the change is test-only and the module it covers is unchanged.

Checklist

• pnpm check:ci and pnpm test both pass (plus cargo fmt / clippy / test for Rust changes)
• I ran the affected surface and observed the change working — the vitest project, including the two mutation runs above
• If a wire message changed: WIRE_PROTOCOL_VERSION is bumped (not applicable)
• New code and assets are my own work, or their origin and license compatibility are noted above
• Docs and comments are updated where behavior changed (no behavior change)

[linear-code]

CODE-478