[Fix] Compare TydiQA answers case-insensitively

[MohammadHijjawi97]

Motivation

TydiQAEvaluator.score lowercases each prediction before comparing it, but the gold answers are compared as-is, since general_postprocess does not lowercase. So a gold answer containing an uppercase letter can never reach exact match, and F1 drops every capitalized token, even when the model reproduces the reference verbatim:

TydiQAEvaluator().score(['Paris'], [['Paris']])
# {'exact_match': 0.0, 'f1': 0.0}

This affects all TyDiQA-GoldP languages with letter case (English, Russian, Finnish, Indonesian, Swahili, ...). The SQuAD v1.1 script this evaluator is adapted from lowercases both sides.

Modification

Lowercase the reference answers alongside the prediction in TydiQAEvaluator.score. Add tests/datasets/test_tydiqa.py, covering exact match / F1 with capitalized references and a wrong answer.

BC-breaking (Optional)

No API change. TyDiQA scores for cased languages will go up, since correct answers are no longer rejected because of case.

Use cases (Optional)

N/A

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues. (flake8, isort, yapf and codespell on the changed files)
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects.
• CLA has been signed and all committers have signed the CLA in this PR.