Placing wide characters into MIME::Entity objects can lead to
double-encoding, as discovered most recently in d469cacc.  Explicitly
decode all headers as UTF-8 when retrieving them with ->get(), and
encode them as UTF-8 before updating them with ->set() or ->replace().
This also applies to headers passed to ->build().  The only exceptions
to this are fixed strings in the source (which, in the absence of "use
utf8", are always bytes).

While the majority of these headers will never have wide characters in
them, always decoding and encoding ensures the proper disipline to
guarantee that strings with the "UTF8" flag do not get placed in a
header, which can cause double-encoding.