[New Model] ProSST (NeurIPS2024)

[tyang816]

Hi, Pascal

My vscode seems to have automatically closed the previous PR... but this is not a big problem.

I added structure tokenization in the new code according to @fteufel. And our paper ProSST: Protein Language Modeling with Quantized Structure and Disentangled Attention has been accepted by NeurIPS 2024.

My previous PR:

We recently developed a new structure-sequence model called **ProSST** which gets a weighted average score of **0.504** at K=2048 on the substitution benchmark. 

1. Added the scoring code of *prosst* in `proteingym/baselines/prosst`.
2. Added *prosst* information to `config.json` and `constants.json`.
3. Added scoring method and [Huggingface Path](https://huggingface.co/AI4Protein) in scripts, `scripts/scoring_DMS_zero_shot/scoring_ProSST_substitutions.sh`.

It is worth noting:
1. A new conda environment is required, the configuration file is in `proteingym/baselines/prosst/environment.yaml` (same as **ProtSSN**).
2. The original PDB files are same with **ProtSSN**, you can [download](https://lianglab.sjtu.edu.cn/files/ProtSSN-2024/proteingym_v1_pdb.tar.gz) them here. The processed sequence and structure files can be [downloaded](https://drive.google.com/file/d/1lSckfPlx7FhzK1FX7EtmmXUOrdiMRerY/view?usp=sharing) here.
3. More model information can be seen in [ProSST repository](https://github.com/ai4protein/ProSST).

Thanks again for your community efforts, and we will continue to pay attention to your group's work.

[pascalnotin]

@tyang816 -- thank you for the updates, and congrats on your acceptance at NeurIPS!
@fteufel has been looking into your PR (thanks again!) and we should be able to merge it in the main branch soon.
Best regards,
Pascal