QXmlStreamReader::addData(QASV) overload unconditionally converts UTF-16 and L1 to UTF-8

The addData(QAnyStringView) overload unconditionally converts Latin1 and UTF-16 strings to UTF-8. This can be fine if it's the first data added to the reader, or if the previous data was also in the UTF-8 encoding (or was converted to it).

However, assume a case when we originally provided some raw data (via a constructor taking QByteArray or addData(QByteArray) overload), and this raw data had a proper XML prolog with an "encoding" attribute set. The specified encoding could be one of Latin1 or UTF-16, in which case the parser would internally use the respective decoder.

After that, appending a string in the proper encoding should just work fine. However, in practice it will result in corrupted data (for Latin1) or even a parsing error (for UTF-16), because we unconditionally convert this data to UTF-8.

This can be illustrated by a simple example:

const QString utf16Str = u"<?xml version=\"1.0\" encoding=\"utf-16\"?>"
                          "<foo><a>Some Data</a>"_s;
// as byte array
const QByteArray utf16Data{
    reinterpret_cast<const char *>(utf16Str.utf16()),
    utf16Str.size() * 2
};

QXmlStreamReader reader(utf16Data);
bool res = reader.readNextStartElement(); // Read <foo>, OK
res = reader.readNextStartElement(); // Read <a>, OK
QString text = reader.readElementText();

// append more UTF-16 data
reader.addData(u"<a>Other Data</a>"_s);
res = reader.readNextStartElement(); // Read <a>, FAIL!
qDebug() << reader.errorString(); // "Premature end of document."

The problem is that newly-added data is converted to UTF-8 and has twice less bytes than the UTF-16 parser expects.

For the Latin1 + Latin1 case the parsing will work, but non-ASCII characters will be read incorrectly.