Add larger support for pytorch model hub

[David-Nevezi-Strango]

I am currently facing some issues, regarding trying to put into PIT a predefined model from pytorch hub. Currently i ve faced multiple issues, such as, for annotation.py, line 255, n.meta has no key "tensor_meta" for example

[matteorisso]

Hello dear @David-Nevezi-Strango, can you provide a minimum reproducible example so that I can better investigate your problem?

[David-Nevezi-Strango]

Yes, sorry about not leaving more details about it

so the PIT constructor fails if i call it using the following lines (you may try different networks as well, for some it will write unsupported OP such as adaptive avgpool2d, for others it will give KeyError: "tensor_meta"):
model_pt = torch.hub.load('pytorch/vision:v0.10.0', 'resnet18', weights=tv.models.ResNet18_Weights.DEFAULT)
model = PIT(model_pt, input_shape=input_shape, cost=params, discrete_cost=True)
where input_shape is given by the dataset and params is from from plinio.cost import params

Torch version is 2.3.0 with torchvision 0.18.0

EDIT: checked torch version again, it is 2.3.0