Optimize Couchstore reads

We'd like 2 related of optimizations to make reads from Couchkvstore faster:

The first is have the new update seq number returned back to the caller, probably by updating the docInfo structure that's passed by the caller. The caller can then stash that with the doc meta info.

The second is to do doc lookups using that seq number, in batch mode. This will require adding a new public api to take an array of docInfos and load them in bulk via by_seq btree.

This task is required by MB-5164 in ep-engine