[quant][core][gpu][improvement] Set tensors as virtual in quantized cudnn linear op
Summary:
For cudnn, if certain intermediate tensors are not needed for other
parts of the operation graph in cudnn, the it can be set as virtual, so
that a pointer-uid pair does not need to be provided for the variant
pack, allowing for better performance.
Test plan:
```
python test/test_quantization.py -k test_qlinear_cudnn
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/75967
Approved by: https://github.com/jerryzh168
