So essentially, you'd have to:
- - encode to (shortest possible) UTF-8
- - apply base64 encoding
- - prepend "=?utf-8?b?"
- - append "?="
*canonicalized/normalized*, UTF-8, not shortest possible UTF-8.
and then you'd have a different encoding for the local-part
than for the domain (which will surely be IDNA).
and it still would break in the face of agents that decode
and/or translate encoded-words.
it's just not going to happen this way.
Keith