GraphQL API perf optimizations

[peferron]

This PR optimizes the performance of fetching cubes via GraphQL.

• Extends the existing optimization for name to sibling fields. That's the largest gain; around 2x faster when running locally against a copy of our database. I didn't measure a significant perf regression for name-only use cases.
• Adds an index to node.type, which is useful when fetching cubes since they only represent a small fraction of the overall nodes.
• Avoids a DB round-trip & materializing a potentially long list of node IDs by using an EXISTS subquery instead, which the Postgres planner will optimize as needed.
• Remove an apparently obsolete LIKE '%_DOT_% filter which couldn't use indexes—I'll need confirmation that this is indeed safe to remove.

Also adds support for @include and @skip directives; before clients would have to alter the query string itself to get the expected perf gains, which is inconvenient.

Finally, fixes an existing bug (happening on main) where metrics and dimensions within revisions are empty unless non-name fields are also selected. For example, the query below returns empty metrics and dimensions, unless you uncomment some of the fields:

{
  findNodes(nodeTypes: [CUBE], limit: 1) {
    revisions {
      cubeMetrics {
        name
        # columns {
        #   name
        # }
      }
      cubeDimensions {
        name
        # attribute
      }
    }
  }
}

[netlify]

✅ Deploy Preview for thriving-cassata-78ae72 canceled.

Name	Link
🔨 Latest commit	d173030
🔍 Latest deploy log	https://app.netlify.com/projects/thriving-cassata-78ae72/deploys/6a2b54817dd4c700085608f6

[shangyian]

Thanks for adding this @peferron!

Remove an apparently obsolete LIKE '%DOT% filter which couldn't use indexes—I'll need confirmation that this is indeed safe to remove.

I just ran some confirmation queries in Postgres and this looks like a reasonable assumption:

dj=> SELECT cr.cube_id, col.name, col.node_revision_id, nr.type
FROM cube cr
JOIN "column" col       ON col.id = cr.cube_element_id
LEFT JOIN noderevision nr ON nr.id = col.node_revision_id
WHERE (col.name LIKE '%\_DOT\_%' ESCAPE '\')
   <> (col.node_revision_id IS NOT NULL AND nr.type = 'METRIC')
LIMIT 50;
 cube_id | name | node_revision_id | type
---------+------+------------------+------
(0 rows)