Add multilingual benchmarking

[victorigualada]

Description

This PR adds multilingual benchmark support.
There are a few decisions that might be questionable but I think it's a simple first approach.
It adds a first multilingual benchmark for devstral-2512.

1. I decided to create assist-<language> directories for each dataset language so we don't change the existing structure. We can have a follow-up PR that will create a tree structure:

|- assist
  |- en
  |- es
  |- fr
|- assist-mini
  |- en
  |- es
  |- fr
...

2. The current leaderboard stays as-is with english only results. Then for each dataset (assist, assist-mini, questions and automations) a new section is created with a table having models as rows and languages as columns. Again we can think on how to repurpose this but as groundwork I think it's good enough.

Let me know what do you think and if we can improve it.

[allenporter]

Generally seems fine though i saw that there are two CI failures
(1) there is a codespell error
(2) i think some snapshots are failing

[allenporter]

Not sure if its super useful to translate the comments too :)

[allenporter]

There are still some test failures. I think there maybe something wrong with the linting configuration in the pre-commit that caused this? May need to manually fix.

[victorigualada]

There are still some test failures. I think there maybe something wrong with the linting configuration in the pre-commit that caused this? May need to manually fix.

Yes that's totally the issue and drove me mad last week. The different linting approach in pre-commit and CI is causing this. I'm thinking if they should be aligned so this doesn't happen again in the future.