PROTOTYPE/WIP: support CDC

[tyler-french]

This PR describes the implementation of CDC for Bazel, using FastCDC 2020 algorithm.

The key design consideration here are:

During a read:

• Check if a blob is above the chunking threshold
• Call SplitBlob to see if the remote has stored the blob with chunking map[blob] -> chunks
• If so, assemble these and use this. Keep the chunks in the disk cache. Blob is downloaded

During a write:

• Check if a blob is above the chunking threshold
• If so, run FastCDC and store the offests + lens of the chunks
• While uploading, we have the files available to upload from the merkle tree. So, instead of uploading the whole blob:
  • Check if this blob -> chunk mapping already exists using Split
  • Call FMB to find the chunks that need uploading (** key optimization, many times the remote has most chunks **)
  • Upload the chunks
  • Call SpliceBlob to register with the server that this blob digest can be constructed using these chunks

Optimization: we don't need to store the individual chunks for uploads, since we have the whole files locally anyway. Storing their offset + len is sufficient to do all of the work.

Depends on a new version release of remote_apis repo.