QStringDecoder::appendToBuffer produces inconsistent results.

Discovered at: https://codereview.qt-project.org/c/qt/qtbase/+/488440/23..24/src/corelib/text/qstringconverter.h#b245

QStringDecoder::appendToBufferchar16_t *out, QByteArrayView ba)

produces inconsistent results when feeding the data piecemeal, compared to feeding the data in one chunk or using alternative decoding facilities as 'QString::fromUtf8()'

The unicode standard doesn't enforce a strict design on how to handle illegal sequences, but I think we can all agree that the one we use should be consistent across all conversion mechanism we provide.

    const char illFormed[] = u8"a\xe0\x9f\x80""a";

In this illegal sequence, we have: 1 code-point (a), \xe0 (legal), \x9f (illegal), \x80 (illegal), 1 code-point(a)

QString and QStringDecoder will write a replacement character for every illegal sequence that we encounter:   

     "a\uFFFD\uFFFD\uFFFDa"

however, feeding the data byte-by-byte will produce:

    "a\uFFFDa"

only writing a single replacement char for the whole sequence. Perhaps some internal state handling in QStringDecoder is producing this inconsistency.

Please find the reproducer attached.

 Reproducer output:
"fromUtf8" OK 
"decoder" OK 
"decoderAppendFull" OK 
"decoderAppendPieceMeal" "a�a" != "a���a"