fix(e2e): flaky Playwright tests — Vite cold start race on ARC runners #3917

Problem

E2E tests intermittently fail with ERR_CONNECTION_REFUSED on localhost:8080 when running on ARC self-hosted K8s runners. The Playwright webServer config waits for the port to be available (30s timeout), but on CPU-constrained runners Vite sometimes takes longer to be fully ready.

Symptoms

• auth.signup.e2e.tests.js — ERR_CONNECTION_REFUSED on page.goto('/signup')
• billing.e2e.tests.js — ERR_CONNECTION_REFUSED or timeout waiting for heading
• Always passes on rerun (second attempt has warm Vite cache)
• Only happens on ARC K8s runners, never locally

Current config

// playwright.config.js
webServer: {
  command: `npx vite --port ${port}`,
  port,
  timeout: 30000, // 30s — may be too short on slow runners
}

Possible fixes

1. Increase webServer.timeout to 60s or 90s
2. Add retries: 1 to Playwright config for CI (env-conditional)
3. Pre-build before E2E: run npx vite build or npx vite optimize before tests so deps are pre-bundled
4. Use stdout: 'pipe' + reuseExistingServer with a readiness check on actual HTTP response (not just port binding)

Impact

Non-blocking (rerun always passes), but wastes CI time and creates noise.