Use 'UTF-8' as the `encoding` attribute in `OutStream` class.

This means that `sys.stdout.encoding` returns something useful instead
of raising an AttributeError. 'UTF-8' was chosen because our version of
write actually wants a unicode string (i.e., not `bytes`) and we want
something universal which an encode the entire range of unicode characters.

IPython/zmq/iostream.py

@@ -4,7 +4,7 b' from io import StringIO'
 
 from session import extract_header, Message
 
-from IPython.utils import io, text, encoding
+from IPython.utils import io, text
 from IPython.utils import py3compat
 
 #-----------------------------------------------------------------------------
@@ -23,6 +23,7 b' class OutStream(object):'
     topic=None
 
     def __init__(self, session, pub_socket, name):
+        self.encoding = 'UTF-8'
         self.session = session
         self.pub_socket = pub_socket
         self.name = name
@@ -73,9 +74,8 b' class OutStream(object):'
         else:
             # Make sure that we're handling unicode
             if not isinstance(string, unicode):
-                enc = encoding.DEFAULT_ENCODING
-                string = string.decode(enc, 'replace')
-            
+                string = string.decode(self.encoding, 'replace')
+
             self._buffer.write(string)
             current_time = time.time()
             if self._start <= 0: