Sort out quantized int support

[dsyme]

Torch has good support for quantized ints, we whould make this availble.

[dsyme]

e.g.

torch.quantize_per_channel(input, scales, zero_points, axis, dtype) → Tensor

and the other basic support for quantized tensors in torch::*quant*