Remove instance-level forward on unpatch

[AlexanderDokuchaev]

What does this PR do?

Prevents unexpected instance-level overrides of the forward function after the original method is restored in ModelPatcher

assert "forward" not in model.__dict__
patcher = ModelPatcher(export_config, model)
with patcher:
    pass
assert "forward" not in model.__dict__  # AssertionError

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

• ONNX / ONNX Runtime : @fxmarty, @echarlaix, @JingyaHuang, @michaelbenayoun

[AlexanderDokuchaev]

Example when "forward" in model.__dict__ raise exception

from transformers import AutoModelForImageClassification

import torch

model = AutoModelForImageClassification.from_pretrained("microsoft/resnet-18").cuda()
inputs = torch.rand(10, 3, 224, 224).cuda()
# Set forward like https://github.com/huggingface/optimum/blob/v1.24.0-release/optimum/exporters/onnx/model_patcher.py#L292
setattr(model, "forward", model.forward)
data_parallel = torch.nn.DataParallel(model)
data_parallel(inputs)  # RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!