Add turn started/completed events and correct exit code on error (#4309)

Adds new event for session completed that includes usage. Also ensures
we return 1 on failures.
```
{
  "type": "session.created",
  "session_id": "019987a7-93e7-7b20-9e05-e90060e411ea"
}
{
  "type": "turn.started"
}
...
{
  "type": "turn.completed",
  "usage": {
    "input_tokens": 78913,
    "cached_input_tokens": 65280,
    "output_tokens": 1099
  }
}
```

codex-rs/exec/tests/suite/server_error_exit.rs

@@ -0,0 +1,34 @@
+#![cfg(not(target_os = "windows"))]
+#![allow(clippy::expect_used, clippy::unwrap_used)]
+
+use core_test_support::responses;
+use core_test_support::test_codex_exec::test_codex_exec;
+use wiremock::matchers::any;
+
+/// Verify that when the server reports an error, `codex-exec` exits with a
+/// non-zero status code so automation can detect failures.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn exits_non_zero_when_server_reports_error() -> anyhow::Result<()> {
+    let test = test_codex_exec();
+
+    // Mock a simple Responses API SSE stream that immediately reports a
+    // `response.failed` event with an error message.
+    let server = responses::start_mock_server().await;
+    let body = responses::sse(vec![serde_json::json!({
+        "type": "response.failed",
+        "response": {
+            "id": "resp_err_1",
+            "error": {"code": "rate_limit_exceeded", "message": "synthetic server error"}
+        }
+    })]);
+    responses::mount_sse_once(&server, any(), body).await;
+
+    test.cmd_with_server(&server)
+        .arg("--skip-git-repo-check")
+        .arg("tell me something")
+        .arg("--experimental-json")
+        .assert()
+        .code(1);
+
+    Ok(())
+}