Build an S3 processing runner

[kylebd99]

Much of the code in the scripts files is devoted to handling S3 and parallelism. It does a few things:

1. Downloading large batches of files using either multiprocessing or s5cmd
2. Applying a function to each batch of files (PDF, metadata json, embeddings, etc.), potentially using multiprocessing to do so
3. Saving results in S3 & checkpointing

We should create a class that handles this logic, so that we can avoid duplicating it every time we want to do something.