python: use surrogateescape when encoding args in wrapper.py

Problem: When passing commandline arguments (i.e. sys.argv) from
Python scripts to C API functions via the wrapper.py class, encoding
errors are possible if a utf-8 string is supplied in argv if the
current environment does not have a UTF-8 LANG or LC_ALL environemnt
set. This seems to be because Python encodes sys.argv to `str` with
the current encoding, and (at least as of Python 3.6) there is no
way to get the actual bytes given in argv in this environment without
throwing a UnicodeEncode error.

Fortunately this has been a big enough problem that PEP383[1] was
devised, introducing the 'surrogatescape' error handler for encode(),
which allows the so-called "smuggling" bytes in character strings.

Since the alternative is to possibly throw UnicodeEncode errors,
add `errors='surrogateescape'` to the wrapper.py __call__() method,
when it encodes string arguments to pass into C. There is likely no
downside here, since there is presumably no existing usage depending
on the current behavior.

[1] https://www.python.org/dev/peps/pep-0383/

src/bindings/python/flux/wrapper.py

@@ -190,7 +190,7 @@ def __call__(self, calling_object, *args_in):
                 # Unpack wrapper objects
                 args[i] = args[i].handle
             elif isinstance(args[i], six.text_type):
-                args[i] = args[i].encode("utf-8")
+                args[i] = args[i].encode("utf-8", errors="surrogateescape")
 
         try:
             result = self.fun(*args)