[DO NOT MERGE UNTIL TOLD] add multi-modal-evals

[mahjongmen]

This PR adds a table to the public_embedding_benchmarks/multi_modal_evals.md file, presenting the performance of various models on different benchmarks:

• The table includes models such as clip-vit-base-patch16, nomic-embed-vision-v1.5, and embed-multilingual-v3.0.
• It provides their respective dimensions and performance metrics on benchmarks like BEIR Average, Coco (2017), Flickr1k, and more.
• The "Total Average" column offers an aggregated view of the models' performance.
• A note below the table specifies that all evaluations were calculated using Recall@10, and certain benchmarks are composites of evaluations gathered and annotated by Cohere.