[Bug]: test runner silently drops the 'screen' key of device descriptors

Version

1.63.0 (code matches current main)

Steps to reproduce

Same device descriptor, two paths, Chromium both times:

Library mode — works:

const context = await browser.newContext({ ...devices['iPhone 13'] });
// window.screen -> { width: 390, height: 844 }  ✅ matches descriptor

Test runner — screen is dropped:

test.use({ ...devices['iPhone 13'] });
test('screen', async ({ page }) => {
  // window.screen -> { width: 390, height: 664 }  ❌ collapses to viewport
});

Control: test.use({ contextOptions: { screen: { width: 390, height: 844 } } }) produces the correct window.screen — the machinery under the runner supports it; only the option plumbing is missing.

Expected behavior

test.use({ ...devices[...] }) forwards all descriptor keys. Six of the seven are registered as TestOptions fixtures in packages/playwright/src/index.ts; screen is the only one with no fixture, no entry in the _combinedContextOptions destructure, and no if-block — and unknown use keys are silently ignored, so nothing warns.

Actual behavior

window.screen (and device-width/height media queries) reflect the viewport instead of the descriptor's screen in test-runner mode.

Additional context

Three signs this is drift rather than design: docs/src/emulation.md promises the runner simulates "screenSize"; the TestOptions.reuseContext docs (v1.62) explicitly list screen among context options that may differ between tests — referencing a test option that doesn't exist; and paramsThatAllowContextReuse in browserContext.ts already includes 'screen'. screen emulation landed in #5967 (v1.10) and the TestOptions list never picked it up.

I have a fix drafted and verified (fixture + destructure + if-chain in index.ts, the overrides-test.d.ts type, docs entry, and a runInlineTest-style regression test — red without the src change, green with it; full playwright.config.spec.ts suite passes). Per the contribution policy I'm filing this first — happy to submit the PR if this is approved for community contribution. Prepared with AI assistance (Claude); the repros and red/green runs were verified by execution.