why was MJPEG quantization support removed

[NickNYF]

I noticed that MJPEG quantization support (values with any quantization) was introduced in issue #566 and further addressed in issue #583 .

However, in issue #944 , it appears that this implementation was removed.

Could you clarify the reason for removing this functionality?
Was it an intentional design decision, or was it removed due to compatibility or implementation concerns?

[NickNYF]

The current implementation incorrectly restricts JPEG quantization table length to only 64 or 128 bytes.
However, in JPEG (DQT segment), the length field represents N × 64 bytes, where N can be any positive integer.
This means valid values such as 192, 256, etc. are also expected in real-world JPEG streams.

The current switch-case effectively rejects valid JPEG quantization tables and breaks RFC2435 / baseline JPEG compatibility.

func (h *headerQuantizationTable) unmarshal(byts []byte) (int, error) {
	if len(byts) < 4 {
		return 0, fmt.Errorf("buffer is too short")
	}

	h.MBZ = byts[0]
	h.Precision = byts[1]
	if h.Precision != 0 {
		return 0, fmt.Errorf("precision %d is not supported", h.Precision)
	}

	length := int(byts[2])<<8 | int(byts[3])
	switch length {
	case 64, 128:  // NOTICE！！ 
	default:
		return 0, fmt.Errorf("table length %d is not supported", length)
	}

	if (len(byts) - 4) < length {
		return 0, fmt.Errorf("buffer is too short")
	}

	tableCount := length / 64
	h.Tables = make([][]byte, tableCount)
	n := 0

	for i := range tableCount {
		h.Tables[i] = byts[4+n : 4+64+n]
		n += 64
	}

	return 4 + length, nil
}