Improve push down predicates

[dennyglee]

Make better use of DocumentDB's native capabilities (e.g. aggregations, ORDER BY, LIMIT, etc.) so a more optimized dataset is returned to Apache Spark. For example:

• For cumulative aggregations, DocumentDB should return those aggregations and Spark should then aggregate accordingly.
• For ORDER BY, DocumentDB can return the data ordered
• For LIMIT, the connector should execute a TOP query to DocumentDB