MEET processed peptide design datasets

Processed train, validation, and test datasets for Scalable Peptide Design via Memory-Efficient Equivariant Transformer.

The archives are expected by the released MEET configs under ./dataset after extraction and include the 100K training split, 1.2M training split, validation split, and test split.

Note that due to the large size of the 1.2M training set, the original archive has been split into 12 shards. To reconstruct the full training archive, download all `train_1p2M.tar.gz.partNN` files into the same directory, then run:

```

cat train_1p2M.tar.gz.part* > train_1p2M.tar.gz

shasum -a 256 -c train_1p2M.tar.gz.parts.sha256

tar -xzf train_1p2M.tar.gz

```