Find a solution allowing empty results

[KennethEnevoldsen]

Some models such as the "translate and embed" models can't be used for cross-lingual tasks, ideally their scores should just be nan.

I am unsure what the best solution is. We could naturally raise an error (but then it seems like it failed). We could also return nan (but then we would have to handle it in mteb, which is also

Seems like the best solution is to raise a specific error something like ModelNotValidForThisTaskError

[KennethEnevoldsen]

@x-tabdeveloping, do you have a suggestion here?