Fix issue with test_unicode test_raiseMemError. The test case now use ``test.support.calcobjsize`` to calculate size of PyUnicode structs. :func:`sys.getsizeof` may return different size when string has UTF-8 memory.