Run benchmarking with the supported models on BEIR MSMARCO

We need to run benchmarking on the [BEIR MSMARCO](https://paperswithcode.com/dataset/beir) dataset, to have a better understanding of how the models are performing for retrieval tasks.

We can use the `test` split available on Hugging Face hub:

[QRels](https://huggingface.co/datasets/BeIR/msmarco-qrels)
[Corpus](https://huggingface.co/datasets/BeIR/msmarco)

Proposed metrics:

- NDCG@10
- Precision@10
- Recall@100

Considering non-judged documents as non-relevant.

Start by reviewing the repository's existing benchmarking entry point and supported-model list, then use the BEIR MSMARCO test split from the linked Hugging Face QRels and Corpus datasets. The work is done when benchmarking results report NDCG@10, Precision@10, and Recall@100, treating non-judged documents as non-relevant.

Written by the indexing model from the issue text.