[Feat] Support dequantization of an exported model

Context

As models heads are not always quantized, it could be useful to export deQuant operators at the end of a quantized model.