Harness primitive for testing expected crashes / uncaught exceptions

まず test_exception/testFinalizerException.js と test_fatal を読んで、サブプロセスに対するそれぞれの期待値を比較します。提案された expectCrash の形を中心に、実装者が提供する harness を設計し、stderr のマッチングと、例外またはシグナルに基づく終了を含めます。両方の対象テストで helper を使用でき、期待される crash の結果を引き続き検証できれば完了です。

Problem

Several upstream Node.js tests verify behavior that causes the process to exit abnormally — uncaught exceptions from finalizers, fatal errors, etc. These tests use a subprocess pattern:

1. Spawn a child process that loads the addon and triggers the crash
2. Assert on the child's exit code and stderr output

The CTS currently has no equivalent harness primitive for this pattern.

Affected tests

• test_exception/testFinalizerException.js (js-native-api) — finalizer throws during GC, expects process exit with "Error during Finalize" on stderr
• test_fatal (node-api) — calls napi_fatal_error, expects process abort with specific message

Proposed solution

Add a harness helper that runs a code snippet in a subprocess and asserts on the outcome:

// Possible API shape:
await expectCrash({
  code: () => {
    const addon = loadAddon('test_exception');
    addon.createExternal();
    // trigger GC...
  },
  stderr: /Error during Finalize/,
  exitCode: (code) => code !== 0,
});

Each implementor would provide the subprocess execution mechanism (e.g., Node.js would use child_process.spawnSync).

Considerations

• The helper needs to be implementor-provided since subprocess APIs are runtime-specific
• The test code to run in the subprocess may need access to loadAddon and other CTS globals
• Some crashes are signal-based (SIGABRT from napi_fatal_error) vs exception-based — the helper should handle both