[quant][graphmode] Support ops with fixed quantization parameters (#38278)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38278
Support ops like aten::hardsigmoid that has a fixed quantization parameters:
```
constexpr float o_scale = 1.0f / 256.0f;
constexpr int32_t o_zero_point = 0;
```
Ops supported:
- hardsigmoid
- sigmoid
- tanh
Test Plan: Imported from OSS
Differential Revision: D21559811
fbshipit-source-id: 26f3c9c3389dea4f07b350172e2974fac8c5c470