WIP: feat: Add ColBERT

[bclavie]

Hey @Muennighoff!

Just the indexing code for now (will add the rest tomorrow), but opening the draft PR in case you wanted to take a look at this before the rest comes in!

Goal of the PR

Add support for ColBERT models, starting with Answer.AI's ColBERT-small via an API Answer will host (discussed with @okhat who's also okay with this being the first ColBERT representative), in order to see how multi-vector models of various sizes fare on this benchmark. The querying mechanism within the API is very simple and lives at AnswerDotAI/mteb_arena_colbert_api.

Changes

• The PR relies on an external API, where the index is hosted and queried, and which will simply return documents. It doesn't change the logic of any existing mechanisms.
• It adds the ColBERT indexing code for full reproducibility
• TODO: It adds the querying mechanism, using API calls to fetch the highest scoring document for a given query.
• TODO: It adds utilities to download the pre-built indexes from Wikipedia to be able to query them locally.