[wip][quant][graphmode] produce reference pattern for binary ops and then rewrite to quantized op (#68229)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68229
This PR makes BinaryOpQuantizeHandler to always produce reference patterns, and we rely on
subgraph_rewriter to rewrite the reference qunatized patterns to quantized ops
Test Plan:
```
python test/test_quantization.py TestQuantizeFx
python test/test_quantization.py TestQuantizeFxOps
```
Imported from OSS
Reviewed By: vkuzo
Differential Revision: D32537714
fbshipit-source-id: 456086b308c4446840d8d37997daa6f8f8068479