s3 decode user-defined metadata with unicode.

Not an answer, but poking around with your example it looks like it's =?UTF-8?B? followed by the bytes of the input encoded in base64, followed by ?= to terminate. However, when if I decode your first example I get extra characters in the output. Your input, encoded as utf8 in Python gives me \xc3\x84M\xc3\x84Z\xc3\x95\xc3\x91 S3, and the output of your example when decoded gives \xc3\x83\xc2\x84M\xc3\x83\xc2\x84Z\xc3\x83\xc2\x95\xc3\x83\xc2\x91 S3. The 0xc2 0x84 pair appear to be a control character, which may be present in your input and stripped out by rePost? Hope that helps.

Initially I also came to this conclusion, I also tried to determine the encoding with chardet, but for some reason it also answers utf8. After that I found that it's look very similar to mime encoding, but It didn't return the original string. https://docs.python.org/3/library/email.header.html#email.header.decode_header