Right now BulkWriteReplyItem uses the write_ops::Upserted type to expose information about upserts. 

However, the Upserted type stores both an index as well as an _id. This makes sense in the context of the update command, since the Upserted type there is a top-level field on the command response and users need a way to map the upsert info back to the input operations. 

But in bulkWrite, we are returning this info in the per-statement responses, so it is redundant to include the index. We should just expose the upsertedId instead.