Fix Long Term Quant Testing (#78)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/78

Long term Quant seems to be using their custom graph capture before running prepare and convert. Let us mirror this so we are testing with the proper quant flow

https://fb.workplace.com/groups/257735836456307/permalink/545316467698241/

The change here would be that Quantize2 would run on the stage before Export.

Testing wise the stages are as follows

```
export.capture_pre_autograd_graph --> prepare --> convert --> exir.capture()

|--------------------------------------------------------|    |-------------|
                           Quantize2(stage)                    Export(stage)
```

Reviewed By: digantdesai

Differential Revision: D48488929

fbshipit-source-id: dac012242c18ea9cc39e6d44bb35aca943cf8f2b