[quant][core][gpu][improvement] Removed conv_output and set output tensors as virtual in quantized cudnn conv2d op
Summary:
With support for virtual tensors in cudnn, we no longer have to allocate
conv_output.
Test plan:
```
python test/test_quantization.py -k test_qconv2d_cudnn
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/76787
Approved by: https://github.com/jerryzh168
