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.