Fix DeBERTa token_type_ids

[deutschmn]

What does this PR do?

This PR fixes #15735. It changes the behavior of DebertaTokenizer and DebertaTokenizerFast when passing pair inputs. Before, the token type IDs were all 0. This PR changes this so that the token_type_ids for the tokens of the second sentence are 1.

It also adds a test case to test this behavior (DebertaTokenizationTest.test_token_type_ids). Failed before, passes now.

Before submitting

• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case. DebertaTokenizer always assigns token type ID 0 #15735
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@LysandreJik?

[HuggingFaceDocBuilderDev]

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

[LysandreJik]

Looks good to me, important bugfix. WDYT @SaulLu?

[SaulLu]

Thank you very much for the proposed fix @deutschmn 🤗 !

To have more context, I have traced the history of the changes concerning DeBERTa's token_type_ids:

1. In the first PR #5929 where Deverta was added the ids were 0 for the first sentence and 1 for the second;
2. When the tokenizer fast was added in PR #11387 the choice was at that time to assign the 2 sentences to id 0 (see lines here and here);
3. In a later bug fix in PR #10703, it seems that the ids were aligned to the fast version, i.e. only 0s (diff here - thanks @daniel-ziegler).

Looking at the history I also think it's a bug (nothing mentions that the ids should have been all assigned to 0) and the proposed fix seems the right one!

[deutschmn]

Great, thanks for looking into this! I also checked Microsoft's implementation and it looks like they use 1 for sentence B as well 😊