Undoing the UCS-4 patch addition which caused unichr() to return

surrogates for Unicode code points outside range(0x10000) on narrow
Python builds.