test(kilo): add per-test timeout for slow Windows CI provider calls

[claudiusthebot]

Problem

PR #195 fixed HEALTH_TIMEOUT_MS (the beforeAll health-check loop), which extended kilo serve's startup window from 15s → 120s on Windows CI.

That wasn't the only timeout. The individual test provider.list() returns {all, connected, default} with content also times out — at 20000ms (vitest 4.x's default testTimeout):

× provider.list() returns {all, connected, default} with content  20010ms
→ Test timed out in 20000ms.

The kilo server itself starts fine (health check passes in 9ms). But provider.list() makes upstream API calls from inside kilo serve to discover providers — on Windows CI runners, these external round-trips consistently exceed 20s. The provider.auth() test is also at risk (14476ms in the same run — close to the limit).

Root cause

The per-test timeout (testTimeout) is separate from the beforeAll hook timeout. PR #195 only extended the latter. Each slow test needs its own timeout argument.

Fix

Add TEST_TIMEOUT_MS = process.platform === "win32" ? 60_000 : 20_000 and pass it as the third argument to provider.list() and provider.auth() tests — the same pattern opencode-live-discovery.test.ts uses for HEALTH_TIMEOUT_MS.

// Before
it("provider.list() returns {all, connected, default} with content", async () => {
  ...
});

// After
it(
  "provider.list() returns {all, connected, default} with content",
  async () => { ... },
  TEST_TIMEOUT_MS,  // 60s on Windows, 20s elsewhere
);

No behaviour change on Linux/macOS (still 20s, same as vitest default).

Evidence

CI run 25968455011 on PR #167 (triggered after #195 merged):

• Backend Live (kilo, ubuntu-latest) ✅
• Backend Live (kilo, macos-latest) ✅
• Backend Live (kilo, windows-latest) ❌ — provider.list() timed out at exactly 20000ms

Effect

Should unblock PR #167 auto-merge once CI goes green with this fix included on main.