[php-src] PR #23959: ext/mbstring: Don't emit surrogates when encoding to UTF-8

Pull Request: https://github.com/php/php-src/pull/23959 Author: nicolas-grekas mb_convert_encoding("\0\0\xD8\0", 'UTF-8', 'UCS-4') returns "\xED\xA0\x80", which isn't valid UTF-8. Since 4427b2e1ab6 (PHP 8.3), mbstring also flags its UTF-8 output as valid, so mb_check_encoding(), mb_scrub(), preg_match('//u') and others accept that string while they reject the same bytes written as a literal. The same happens with UCS-2 and HTML-ENTITIES input, mb_convert_variables(), mb_decode_numericentity() and mb_decode_mimeheader(). The UTF-8 encoder now replaces surrogates with the substitute character, as it already does for codepoints above U+10FFFF. The internal mode used by mb_strpos() and friends keeps encoding them, so that different surrogates in a UCS-2 string still don't match each other. Found while working on #23957.