Backport PR #4668: Make non-ASCII docstring unicode

We shouldn't have non-ascii characters in a bytes literal. This seems to be the source of problems for the docs build on ShiningPanda.

Merging straight away, just making a PR so there's a record of this.

IPython/utils/io.py

@@ -288,7 +288,7 b' class capture_output(object):'
         sys.stderr = self.sys_stderr
 
 def unicode_std_stream(stream='stdout'):
-    """Get a wrapper to write unicode to stdout/stderr as UTF-8.
+    u"""Get a wrapper to write unicode to stdout/stderr as UTF-8.
 
     This ignores environment variables and default encodings, to reliably write
     unicode to stdout or stderr.