Add support for adding non-key columns to indexes

[Hannes Ljungberg]

Postgres got support for the INCLUDE clause in CREATE INDEX. This can be used to add non-key columns to the index.

CREATE INDEX idx
    ON t1 ( col1 )
     INCLUDE ( col2 )

This allows for Index Only Scans on queries like:

SELECT col1, col2 FROM t1 WHERE t1 = 'foo';

More info:
​https://www.postgresql.org/docs/current/sql-createindex.html
​https://use-the-index-luke.com/blog/2019-04/include-columns-in-btree-indexes
​https://www.postgresql.org/docs/current/indexes-index-only-scans.html

The idea is to add an additional kwarg to Index to support this:

Index(
    name='some-idx',
    fields=['headline'],
    include=['pub_date']

One of the biggest possibilities of this feature is to add included columns to unique indexes and use them to perform Index Only Scans. This would require adding the same kwarg to UniqueConstraint. The implementation would be a lot like the condition kwargs to both Index and UniqueConstraint.

At the moment the only Django-supported database that can use this feature is Postgres but it's also supported by Microsoft SQL Server and IBM Db2 with the same syntax. Because of this I think it shouldn't be implemented as a postgres only feature but exposed on BaseDatabaseSchemaEditor to ease the adding of support when/if sqlite or mariadb/mysql get it.

I have a proof of concept I'm working on and will create a PR soon if accepted.