Make AutoFormer work with previous torch version

[ydshieh]

What does this PR do?

Without import torch.utils.checkpoint (which we have in other files, like Bart), with torch 1.13, we got an error

(running RUN_SLOW=1 python3 -m pytest -v tests/models/autoformer/test_modeling_autoformer.py::AutoformerModelTest::test_training_gradient_checkpointing)

>                   layer_outputs = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(encoder_layer),
                        hidden_states,
                        attention_mask,
                        (head_mask[idx] if head_mask is not None else None),
                    )
E                   AttributeError: module 'torch.utils' has no attribute 'checkpoint'

Let's make it work with previous torch version(s) ❤️ .

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[amyeroberts]

Thanks for fixing!