There are many quantized::add APIs: https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/quantized/cpu/qadd.cpp#L253-L260

But we should reduce the numbers so that some of the functions can be reused.
Blocked by quantized::add support broadcasting.

cc @jerryzh168 @jianyuh @dzhulgakov @raghuramank100 @jamesr66a