Make scripts/download_datasets.py download BEIR datasets

[Witiko]

In Dockerfile, we call the script scripts/download_datasets.py that downloads all datasets to /var/tmp/pv211, so that the datasets are shared by all students that use JupyterHub, saving time and disk space. For example, here we download the ARQMath datasets: 1->2. Here, the students load them: 3->4->5.

Since #3, we've supported BEIR datasets. However, the BEIR datasets are not downloaded in Dockerfile and they are saved and loaded from the ./datasets directory, which slows down the students and duplicates disk space occupied.

Tasks

• Download BEIR datasets to /var/tmp/pv211 in scripts/download_datasets.py.
• Load BEIR datasets from /var/tmp/pv211 in pv211_utils.beir.loader.

[Witiko]

Since #1, we've had the pv211_utils.datasets module. However, the ArqmathDataset and TrecDataset classes hardcode the /var/tmp/pv211 path without a way to set it from the outside, which makes it more difficult to use pv211-utils as a general-purpose library outside the confines of the PV211 course.

Bonus tasks

• Add optional cache_directory parameter to pv211_utils.datasets.ArqmathDataset.load_answers.
• Add optional cache_directory parameter to pv211_utils.datasets.ArqmathDataset.load_questions.
• Add optional cache_directory parameter to pv211_utils.datasets.TrecDataset.load_documents.