Ability to filter available testsets by language pair

[ZJaume]

Implement #210

Using

sacrebleu --list -l en-fr

test sets available in English-French are listed.

export SKIP_CHRF="true"; export SKIP_TER="true"; export SKIP_MECAB="true"; export SKIP_MECAB_KO="true"; ./test.sh

tests are passing.

Some examples:

sacrebleu --list -l is-en

The available test sets for is-en are:
wmt21                         : Official evaluation data for WMT21.
wmt21/dev                     : Development data for WMT21，if multiple references are available, the first one is used.

sacrebleu --list -l tr-en

The available test sets for tr-en are:
wmt16                         : Official evaluation data.
wmt16/dev                     : Development sets released for new languages in 2016.
wmt17                         : Official evaluation data.
wmt18                         : Official evaluation data.
wmt18/test-ts                 : Official evaluation sources with extra test sets interleaved.

sacrebleu --list -l fa-en

The available test sets for fa-en are:

Swapping the language pair gives the same output for consistence with download, that is able to find the set regardless of direction I guess.

[ZJaume]

Please let me know if something with codestyle or style of implementation is wrong and I will change it. Hope it is useful!

[ZJaume]

I thought test.sh was running pytest. Now it should be fixed, python -m pytest -k available works for me.

[mjpost]

Thanks! Can you give me write permission to your branch, so I add a few small changes (e.g., I updated the CHANGELOG).

[mjpost]

LGTM, thanks for the PR!

[mjpost]

@ZJaume Let me know if you can give me write access. If so, I'll push up a few changes and merge. If not, I'll have to create a local branch and have someone else approve it.

[ZJaume]

There you go! I gave you collaborator access to my fork.

[mjpost]

Thanks very much! I'll release this with 2.2.2 soon.