This is not actually a duplicate of 11783.  Rereading (parts of) that issue, we decided we currently have no good way to do automatic conversion between unicode and internationalized domains, so the user of the library has to do it themselves.  This means that the bug *here* is that the new email API is *wrongly* encoding the non-ascii in the domain by using an encoded word.  I'm surprised at that; I thought I'd guarded against it.

What should be happening here is that an error should be raised when that header is set (or possibly when it is accessed/serialized, but when set would be better I think) saying that there is non-ascii in the domain part.