Email header refolding in email._header_value_parser adds additional carriage return symbols to the end of nested parse trees, when used with an EmailPolicy with linesep='\r\n'. This leads to broken email headers when composing an email with a "To:" or "CC:" header having a comma-separated list of recipients with some of them containing non-ASCII characters in their DisplayName.

The bug seems to be caused by the following line (in Python 3.7):
`encoded_part = part.fold(policy=policy)[:-1] # strip nl`
(email/_header_value_parser.py, line 2629)

This line calls part.fold() / _refold_parse_tree() recursively and tries to remove the trailing newline, which is added by the recursive call of _refold_parse_tree(). Unfortunately, it fails to do so, if the length of the policy's line separator sequence does not equal 1. Thus, this line should be corrected to something like
`encoded_part = part.fold(policy=policy)[:-len(policy.linesep)] # strip nl`

Thank you for the report.  This is a duplicate of #34424, which has a PR.