When passing a unicode string to smtplib.SMTP.sendmail() as "msg"
argument, there is an implicit bytes encoding using "ascii" encoding in
python3. Of course this does not work if the string contains non-ASCII
characters. In fact, config's sendmails method intent to pass bytes to
smtplib.SMTP.sendmail() as it uses msg.as_string() method.
Unfortunately, in python3, this method returns a unicode string whereas
it returns a bytes string in python2; we thus fix this by calling
as_bytes() method on python3.

As there is no "as_bytes" method in python2, we need to handle python2
compatibility by hand and either call as_string() or as_bytes().

In testlib, where we mock smtplib.SMTP, we need to keep the "msg"
argument of Email class (defined in testlib as well) a unicode string.
Otherwise, it fails to be parsed by email.message_from_string() (from
stdlib) if it is bytes on python3.