This revisits the fixes for #409 (merged in !301), as when
trying to parse unencrypted, but signed utf-8 mails, they were
failing to be parsed and thus generated another error. More
in-depth later.

With this fix we are changing the approach:

  1. We switch to UTF-8 as default input
  2. If this is not a valid encoding, we try to convert the
     input to UTF-8
  3. If we are failing to convert, we scrub the input, so that
     at least what is proper UTF-8 will be passed on.

ASCII-8BIT is a BINARY encoding and the mail library will only
force the mail to have CRLF if it only contains ASCII conent.
While UTF-8 is not a BINARY encoding and thus will get CRLF if
it is a valid encoding. Getting CRLF is important for mails such
as the one in `signed_utf8.eml`, as otherwise the parts detection
will fail and the whole body will end up in the prologue, with no
parts.
Enforcing UTF-8 will still make some of our charset mails failing,
as they are - validly - not UTF-8. By using a new dependency
'charlock_holmes', we are able to detect the actual encoding and
thus try to convert it to UTF-8. If everything fails, we just
drop the invalid characters.