[ROCm] Add miopen determinism support for convolutions (#107028)
With torchvision installed many of the test_distributed_spawn tests failed due to divergence between model runs. To resolve this we are adding the MIOPEN_CONVOLUTION_ATTRIB_DETERMINISTIC attribute to support deterministic convolutions on ROCm.
This means examples such as https://github.com/pytorch/pytorch/blob/main/torch/testing/_internal/distributed/distributed_test.py#L4777 which use the torch.backends.cudnn.flags.deterministic flag will behave correctly on ROCm
```
with torch.backends.cudnn.flags(
enabled=True, deterministic=True, benchmark=False
):
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107028
Approved by: https://github.com/jeffdaily, https://github.com/kit1980
