Add missing type hints for all flavors of RoBERTa PyTorch models.

[chainyo]

What does this PR do?

I added type hints for all RoBERTa PyTorch classes as described in #16059 .
As I commented on the linked issue, CamemBERT inherits from RoBERTa, so it should be also good for PyTorch version of CamemBERT models.

@Rocketknight1

I was not sure about encoder_hidden_states and encoder_attention_mask which have torch.FloatTensor as type in the docstring, but I don't know if it needs to be specified or torch.Tensor is enough. Tell me if I need to change that.

EDIT: Oh it seems that I have put some torch.LongTensor like inputs_ids as torch.Tensor.

[HuggingFaceDocBuilderDev]

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

[Rocketknight1]

@chainyo it's totally okay to just use torch.Tensor, specifying torch.FloatTensor or torch.LongTensor is a bonus but not at all necessary!

Also, this PR looks great, but note that the methods can return a tuple if return_dict is False, so the real return type is usually Union[Tuple, OutputClass] and not just OutputClass!

[chainyo]

@chainyo it's totally okay to just use torch.Tensor, specifying torch.FloatTensor or torch.LongTensor is a bonus but not at all necessary!

Also, this PR looks great, but note that the methods can return a tuple if return_dict is False, so the real return type is usually Union[Tuple, OutputClass] and not just OutputClass!

Fine, I will change that.

[Rocketknight1]

Looks great, thank you!