Semantics of zero-point

From quantized-tensor documentation, the quantized dtypes can be sub ints (like torch.qint8) and zero-point is of type int. I am trying to understand the rationale for this wider zero-point type? For example, is it to express zero point outside the storage type range or it is just a catch all to express zero-point in any supported quantized dtype ? Please let me know any insights on this?