This was regressed by bug 1788115, which allowed us to parse ASCII labels
again to enforce they're valid. However, the existing code only checked for
invalid punycode (that fails decoding), not ACE labels that are invalid.
If we don't check for this condition uidna_labelToUnicode will add a U+FFFD
character to the end of the label, making us encode it again, changing the URL.
When that happens, the principal's origin may change between serializations,
which is not OK.
Depends on D159373
Differential Revision: https://phabricator.services.mozilla.com/D160785
140dd131bc