Try to fix reentrant write transient failures in tests (#5447)

* Disable print_destroyed in tests on GraalPy

* Reenable test_iostream on GraalPy

tests/constructor_stats.h

@@ -312,8 +312,16 @@ void print_created(T *inst, Values &&...values) {
 }
 template <class T, typename... Values>
 void print_destroyed(T *inst, Values &&...values) { // Prints but doesn't store given values
+    /*
+     * On GraalPy, destructors can trigger anywhere and this can cause random
+     * failures in unrelated tests.
+     */
+#if !defined(GRAALVM_PYTHON)
     print_constr_details(inst, "destroyed", values...);
     track_destroyed(inst);
+#else
+    py::detail::silence_unused_warnings(inst, values...);
+#endif
 }
 template <class T, typename... Values>
 void print_values(T *inst, Values &&...values) {

tests/test_iostream.py

@@ -6,14 +6,8 @@ from io import StringIO
 
 import pytest
 
-import env  # noqa: F401
 from pybind11_tests import iostream as m
 
-pytestmark = pytest.mark.skipif(
-    "env.GRAALPY",
-    reason="Delayed prints from finalizers from other tests can end up in the output",
-)
-
 
 def test_captured(capsys):
     msg = "I've been redirected to Python, I hope!"