Provide a consistent order of mixed index results

Elasticsearch search uses Lucene doc id as the tiebreaker, which is different among the same documents in different shards. This means if you have a query (assuming it hits Elasticsearch):

`g.V().has("name", "value").range(0, 10000)`

and a subsequent query

`g.V().has("name", "value").range(10000, 20000)`

there is no guarantee that results are continuous, even if there is no write request in-between at all.

Why don't we simply use `g.V().has("name", "value").range(0, 20000)`? There are a couple of reasons: 1) a single query would take too much time and our web server will simply cancel it 2) ability to fire search queries concurrently 3) avoid using too much memory at a time

We have the need to ensure a series of gremlin queries (like above) return consistent results (we don't have many write requests so we can assume documents remain unchanged in a period of time). To achieve this, we end up putting a custom unique field (ensured by application logic) in each document and patch ES6Compat/ES7Compat to use that field as the default sorting key. So far, so good.

As a general solution, I think it would be reasonable if JanusGraph can optionally document the _id field as a keyword, and uses it as a tie-breaker (sorting key).

This is related to https://github.com/JanusGraph/janusgraph/issues/1680 but from a different aspect. The solution to both might be the same, though.

Start with the ES6Compat and ES7Compat entry points named in the issue, and inspect how Elasticsearch sorting and tie-breaking are configured. Determine how an optional document _id keyword field could provide stable ordering across shards, then identify the relevant coverage needed to confirm paginated ranges remain continuous.

Written by the indexing model from the issue text.