Separate out S3 Object reference keys from search keys

[shyamd]

It's a little ambiguous on how to set up the S3 store to have some subset of search data in the index. This is partly because the keys for creating the S3 Object and the index keys are the same. We should make an init attribute in the S3Store: metadata or something that the user supplies for fields they want to be able to search on. The S3Store should use the keys field to determine uniqueness when it looks for a doc in the index DB to update. This will make it so that the naming of the blob in S3 can just be something like an OID.

@jmmshn @munrojm