* find a better way to run "find" without creating the index storage, if one
  if one is not already readily available (understand cost now, improve after merge)
* improve performance of splitlines (CF)
* think about cost of utf8 list strategy (CF)
* revisit why runicode import str_decode_utf_8_impl needed instead of runicode
  import str_decode_utf_8
* revisit remaining places in win32 where we do utf8.decode('utf-8'), they should work
  directly with utf8 (can be converted via runicode.str_decode_utf_8 as well)
  - rutf8.utf8_encode_mbcs
  - unicodehelper.fsencode
  - _winreg.interp_winreg
* remove 'assert not isinstance(*, unicode)
* add a flag that prevents support for unicode in rpython and enable it in PyPy (CF, Armin)
* convert all realunicode_w to unicode_w after we flush out all old uses of
  unicode_w
* review all uses of W_Unicode.text_w, right now it is exactly W_Unicode.utf8_w. 
  It shoud only return valid utf8 (see 0be26dc39a59 which broke translation on
  win32 and failed tests on linux64). Then we can use it in places like
  _socket.interp_func.getaddrinfo instead of space.encode_unicode_object(w_port,
  'utf-8', 'strict')