ns for fx: add basic testing on cuda (#65593)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65593
Adds test cases that the three Numeric Suite Core APIs work
when the models are on cuda. In particular:
1. create models and move them to cuda
2. add loggers (if applicable)
3. run data through (if applicable)
4. extract results
It works without code changes because a `Logger` object is
created without any device specific objects (they only get
added if a data is passed through). It's good to have this tested.
Test Plan:
```
python test/test_quantization.py TestFXNumericSuiteCoreAPIs.test_extract_weights_cuda
python test/test_quantization.py TestFXNumericSuiteCoreAPIs.test_add_loggers_cuda
python test/test_quantization.py TestFXNumericSuiteCoreAPIs.test_add_shadow_loggers_cuda
```
Imported from OSS
Reviewed By: raghuramank100
Differential Revision: D31160897
fbshipit-source-id: 8eacf164d0496baf2830491200ea721c0f32ac92