Remove unnecessary DISTINCT ON expressions from introspection SQL. (#260)

### What

There are two:

1. selecting only one relation per name
2. selecting only one aggregate function per input type

These are both unnecessary.

This work was inspired by CockroachDB failing to respect ordering within
CTEs.

### How

The first, selecting only one relation per name, was there to avoid
having to deal with multiple relations of the same name in distinct
schemas, but we now support schemas (and prefix all relations with their
schema name), so this is unnecessary and possibly dangerous.

The second, selecting only one aggregate function per input type, is not
necessary because we only support aggregate functions which take a
single argument, and therefore there can only be one per input type
_anyway_, i.e. you cannot have two aggregate functions of the same name
with the same input type(s) and different return types (return type
polymorphism) in PostgreSQL.

We therefore simplify these queries to avoid having to use `DISTINCT ON`
at all.

Co-authored-by: Philip Lykke Carlsen <[email protected]>

Author: SamirTalwar

crates/connectors/ndc-postgres/src/configuration/version2.sql

@@ -62,12 +62,7 @@ WITH
       cl.relkind relation_kind
     FROM
       pg_class cl
-  ),
-  queryable_relations AS
-  (
-    SELECT DISTINCT ON (relation_name) relations.*
-    FROM relations
-    WHERE relation_kind IN
+    WHERE relkind IN
       -- Lots of different types of relations exist, but we're only interested in
       -- the ones that can be queried.
       (
@@ -82,11 +77,6 @@ WITH
         -- c = composite type,
         -- I = partitioned index
       )
-
-    -- Since we will _not_ be grouping by a key we need this to be ordered
-    -- to get deterministic results.
-    -- (Specificically, we do not yet take schemas into account)
-    ORDER BY relation_name, schema_id, relation_kind
   ),
 
   -- Columns are recorded in `pg_attribute`. An 'attribute' is the generic term
@@ -301,78 +291,37 @@ WITH
   -- their schema.
   aggregates AS
   (
-    WITH
-      -- The arguments to an aggregate function is an array of type oids, which
-      -- we want to resolve to an array of type names instead.
-      -- Somewhat awkwardly, this means we have to unnest, join on types, and
-      -- array_agg and group by.
-      aggregate_argument_types AS
-      (
-        SELECT
-          arg.proc_id,
-          array_agg(arg.type_name) AS argument_types
-        FROM
-        (
-          SELECT
-            proc.proc_id,
-            t.type_name
-          FROM
-          (
-            SELECT
-              proc.oid AS proc_id,
-              unnest(proc.proargtypes) AS type_id
-            FROM
-              pg_catalog.pg_proc AS proc
-            WHERE
-              -- We only support single-argument aggregates currently.
-              -- This assertion is important to make here since joining with
-              -- 'types' filter arguments of polymorphic type, and we might
-              -- risk ending up with one argument later.
-              cardinality(proc.proargtypes) = 1
-          )
-          AS proc
-          INNER JOIN
-            scalar_types AS t
-            USING (type_id)
-        )
-        AS arg
-        GROUP BY arg.proc_id
-        HAVING
-          -- We need to check that we still have an argument, since we're
-          -- filtering by our restricted notion of scalar types, which may
-          -- exclude some types (e.g. pseudo-types and array types).
-          cardinality(array_agg(arg.type_name)) = 1
-      )
     SELECT
       proc.oid AS proc_id,
       proc.proname AS proc_name,
       proc.pronamespace AS schema_id,
-      args.argument_types,
+      arg_type.type_name as argument_type,
       ret_type.type_name as return_type
-
       -- Columns that will likely be of interest soon:
       -- proc.proargnames AS argument_names,
 
     FROM
       pg_catalog.pg_proc AS proc
 
-    INNER JOIN aggregate_argument_types
-      AS args
-      ON (proc.oid = args.proc_id)
+    -- fetch the argument type name, discarding any unsupported types
+    INNER JOIN scalar_types AS arg_type
+      ON (arg_type.type_id = proc.proargtypes[0])
 
-    INNER JOIN scalar_types
-      AS ret_type
+    -- fetch the return type name, discarding any unsupported types
+    INNER JOIN scalar_types AS ret_type
       ON (ret_type.type_id = proc.prorettype)
 
-    -- Restrict our scope to only aggregation functions
-    INNER JOIN pg_aggregate
-      ON (pg_aggregate.aggfnoid = proc.oid)
+    -- restrict our scope to only aggregation functions
+    INNER JOIN pg_aggregate AS aggregate
+      ON (aggregate.aggfnoid = proc.oid)
 
     WHERE
-     --  We are only interested in functions:
-     --  * Which are aggregation functions.
-      -- * Which don't take any 'direct' (i.e., non-aggregation) arguments
-      pg_aggregate.aggnumdirectargs = 0
+      -- We are only interested in functions:
+      -- * which take a single input argument
+      -- * which are aggregation functions
+      -- * which don't take any 'direct' (i.e., non-aggregation) arguments
+      proc.pronargs = 1
+      AND aggregate.aggnumdirectargs = 0
 
   ),
 
@@ -849,7 +798,7 @@ FROM
       )
       AS result
     FROM
-      queryable_relations
+      relations
       AS rel
 
     LEFT OUTER JOIN
@@ -1055,10 +1004,9 @@ FROM
         ) AS routines
       FROM
       (
-        -- We only support aggregation functions that take a single argument.
-        SELECT DISTINCT ON (argument_type, proc_name)
+        SELECT
           agg.proc_name,
-          agg.argument_types[1] as argument_type,
+          agg.argument_type,
           agg.return_type
         FROM
           aggregates AS agg