[Keerthana-64]

validate_email re-validates the local part after NFC normalization (the 2.2.0 fix for #142, where U+037E greek question mark becomes a semicolon), but it normalizes ret.local_part, which has already had unnecessary quotes dropped, and re-validates it with quoted_local_part taken from the input. with allow_quoted_local=True, "a\u037eb"@example.com passes the first check as a plain internationalized dot-atom, loses its quotes, becomes a;b under NFC, is accepted by the second check under quoted-string rules, and comes back as normalized a;b@example.com, which validate_email itself rejects. that bare semicolon is the #142 problem: Python's EmailMessage reads a To header of x;victim@example.com as x, <>, and getaddresses(strict=False) splits it into x, an empty entry and victim@example.com, while "x;victim"@example.com stays one address. normalizing unescaped_local_part and taking local_part from the second validate_email_local_part call quotes the result according to its normalized form; ascii_local_part and smtputf8 still come from the original, as discussed in #185. across about 95,000 random quoted and unquoted local parts under four option sets, 3,698 of 140,427 valid results on main had a normalized form that failed validation or changed when validated again, and none do with this change. the only other differences are that a quoted local part which becomes a plain dot-atom only after normalization (< or > plus U+0338) now drops its quotes like other unnecessarily quoted ones, and one that also contains a unicode space is now rejected, since its normalized form already was. the new test case fails on main and passes here; flake8 and the suite pass on 3.10 to 3.15, mypy on 3.10 to 3.14.