Add WMT 2024 test sets

[mjpost]

Closes wmt-conference/wmt24-news-systems#3

[mjpost]

@martinpopel Python 3.8 is EOLed, are you okay if I drop it from testing? We have a lot of long-running jobs that go and this would help reduce it.

[martinpopel]

Yes, let's drop Python 3.8. However, we should add Python 3.13.
If we need to (further) reduce the test time:

• We could drop Python 3.9 or Python 3.10 (the motivation for the latter is that it seems less probable to encounter a bug that would show only in 3.10, but not in 3.9 and 3.11).
• ./test.sh takes about 11 minutes. For testing sacrebleu -t wmt16,wmt17, we need to download these two tarballs - we could choose smaller ones. We could also prevent downloading https://github.com/MicrosoftTranslator/Translator-HumanParityData/archive/master.zip and https://data.statmt.org/wmt17/translation-task/test-update-1.tgz. That said, I would like to keep at least some real-world test sets for multiple languages in the testing.

[zouharvi]

Hi, feel free to replace the link with this one, which only contains the XML and is much smaller: https://github.com/wmt-conference/wmt24-news-systems/releases/download/v1.1/data_onlyxml.tar.gz