Support simple query pushdown for Postgres

[seeforschauer]

Support simple query pushdown

The goal of this task is to enable simple query pushdown for querier that look like {_id: <ObjectID>}.
Only that field name (_id) is supported, and only that value type (ObjectID).

In the future, we will add support for other fields (starting with simple scalar fields),
other values (starting with strings, numbers, and other simple scalar values),
other conditions like {_id: <ObjectID>, field: value} (so only the first part is pushed down).

Postgres

If a field name is different, value type is not ObjectID, or some other condition is present,
we should use the previous version of the code that SELECTs the whole table without any WHERE condition.

If the value type is ObjectID but the data in the value is somehow corrupted, raise an error (fjson unmarshal).

If those conditions are met, we send a SELECT query with a WHERE condition.

Proof of concept for a {_id: <ObjectID>} pushdown query, PostgreSQL:

GIN index is not used by PostgreSQL in any queries below.
So let's use the first one just because it is first.

-- this will not use these indexes:
-- CREATE INDEX values_id_idx ON public."values" USING gin ((((_jsonb -> '_id'::text) -> '$o'::text)))
-- CREATE INDEX values_id_idx ON values (((_jsonb ->> '_id')::text));
select * from values where (_jsonb->'_id')::jsonb->>'$o' = '507f1f77bcf86cd799439011'; -- no, seq scan will it use index? not that indexes

Test

Integration tests: Provide a new test flag that enables a query pushdown.
Run with and without flag: results must be the same.
When explain feature will be ready, compare the query plan.

Links

Analysis task: #847

[FranckPachot]

GIN index is not used by PostgreSQL in any queries below.
So let's use the first one just because it is first.

I'm not sure I understand this but, just in case, the following index works in PostgreSQL and YugabyteDB:

yugabyte=# create table values (_jsonb jsonb);
CREATE TABLE
yugabyte=# CREATE INDEX values_id_idx ON values (((_jsonb -> '_id'::text) ->> '$o'::text));
CREATE INDEX
yugabyte=# explain
yugabyte-# select * from values where (_jsonb->'_id')::jsonb->>'$o' = '507f1f77bcf86cd799439011';
                                         QUERY PLAN
---------------------------------------------------------------------------------------------
 Index Scan using values_id_idx on "values"  (cost=0.00..5.27 rows=10 width=32)
   Index Cond: (((_jsonb -> '_id'::text) ->> '$o'::text) = '507f1f77bcf86cd799439011'::text)
(2 rows)

It is not a GIN index because it is a single value. But the indexed expression must match the one used in the equality predicate, like here.