Drop special macOS handling for output streams

we had something somewhat similar in e-antic where we strangely needed
to add an explicit ends() to prevent a segfault. However, str() should
create a copy of the string buffer so there should not be any issues
here. (And I don't understand anymore why the problem was fixed by the
ends in e-antic.)

pyexactreal/src/pyexactreal/cppyy_exactreal.py

@@ -381,15 +381,10 @@ def __repr__(self):
                 term<exactreal::Arb ... &>[=1.00000]
 
         """
-        # With cppyy 2.4.2 from conda-forge, the YAP printing below sometimes
-        # segfaults on macOS. We therefore disable a test for it, see arf.py.
         import cppyy
         cppyy.include("boost/yap/print.hpp")
-        os = cppyy.gbl.std.stringstream()
-        try:
-            cppyy.gbl.boost.yap.print(os, self.value)
-        except:
-            return repr(self.value)
+        os = cppyy.gbl.std.ostringstream()
+        cppyy.gbl.boost.yap.print(os, self.value)
         return os.str().strip()
 
 def makeModule(traits, gens, ring=None):

pyexactreal/test/arf.py

@@ -31,12 +31,8 @@ def test_arithmetic():
     a = Arf(1)
     b = Arf(2)
 
-    # The following test segfaults on macOS with cppyy 2.4.2 (we don't
-    # understand why at the moment so we disable the test.)
-    import platform
-    if platform.system() != "Darwin":
-        # Expressions do not get evaluated without specifying a precision/rounding
-        assert str(a + a).startswith("expr<+>")
+    # Expressions do not get evaluated without specifying a precision/rounding
+    assert str(a + a).startswith("expr<+>")
 
     assert (a + a)(64, Arf.Round.DOWN) == b