test(e2e): fail the codex compaction tests when either turn exits nonzero

A first turn that read the secret and then errored could still pass, because the resumed turn
could recall the value from the compacted history. Both compaction tests now require each turn
to exit 0

tests/e2e/codex/test_auto_compaction_e2e.py

@@ -26,6 +26,7 @@ class TestCodexAutoCompaction:
             cwd=workspace,
             compact_after_tokens=COMPACT_AFTER_TOKENS,
         )
+        assert first.exit_code == 0, first
         assert first.thread_id is not None, first
         (workspace / "secret.txt").unlink()
         second: Final = codex.resume(
@@ -35,5 +36,6 @@ class TestCodexAutoCompaction:
             cwd=workspace,
             compact_after_tokens=COMPACT_AFTER_TOKENS,
         )
+        assert second.exit_code == 0, second
         assert any(compaction.remote for compaction in codex.compactions(first.thread_id)), (first, second)
         assert marker in second.answer, second

tests/e2e/codex/test_real_model_name_e2e.py

@@ -42,6 +42,7 @@ class TestCodexUnderTheRealModelName:
             cwd=workspace,
             compact_after_tokens=COMPACT_AFTER_TOKENS,
         )
+        assert first.exit_code == 0, first
         assert first.thread_id is not None, first
         (workspace / "secret.txt").unlink()
         second: Final = real_name_codex.resume(
@@ -51,5 +52,6 @@ class TestCodexUnderTheRealModelName:
             cwd=workspace,
             compact_after_tokens=COMPACT_AFTER_TOKENS,
         )
+        assert second.exit_code == 0, second
         assert any(compaction.remote for compaction in real_name_codex.compactions(first.thread_id)), (first, second)
         assert marker in second.answer, second