[mieb] Task filtering by modality supported by models

[isaac-chung]

Fixes #1632

• Added ModelMeta.modalities, which defaults to ["text"] to be backwards compatible. VLMs will have ["image","text"].
• Added logic in MTEB.run for exact match between task and model modalities
• Added modality model meta to relevant models

Here is the new log, showing that moco only supports images.

## Evaluating 1 tasks:
Any2AnyRetrieval
    - VidoreSyntheticDocQAAIRetrieval, t2i


INFO:mteb.evaluation.MTEB:

********************** Evaluating VidoreSyntheticDocQAAIRetrieval **********************
INFO:mteb.evaluation.MTEB:nyu-visionx/moco-v3-vit-b only supports ['image'], but the task modalities are ['text', 'image'].

Checklist

• Run tests locally to make sure nothing is broken using make test.
• Run the formatter to format the code using make lint.

cc @Muennighoff @gowitheflow-1998

[KennethEnevoldsen]

I Would imagine the ideal message to be:

INFO:mteb.evaluation.MTEB:nyu-visionx/moco-v3-vit-b only supports the modalities ["images"], but the task category is t2i and include both the modalities ["text", "image"].

[isaac-chung]

@KennethEnevoldsen thanks but the PR was still in draft, and we had the same thought! It should be ready now.

[isaac-chung]

Thanks for the quick reviews folks. Merry Christmas! 🎄🎁