Database schema introspection incomplete/empty (PK/FK missing in playground/db-schema)

## Description

I’m trying to extract the database schema from a Postgres instance connected to Cube (CubeJS), but none of the approaches I tried work correctly.

My goal is to reliably retrieve tables/columns plus **primary keys and foreign keys**, so I can generate **Data Models programmatically** (or at least validate/derive relationships). However:

### 1) `playground/db-schema` endpoint

I tried the Playground endpoint `playground/db-schema` (I understand this might not be intended for production use), but the output does **not correctly include primary keys or foreign keys**. Relationships appear incomplete/incorrect.

### 2) Querying `information_schema` via CubeSQL API

I tried using the CubeSQL API endpoint:

```json
POST /cubejs-api/v1/cubesql
{
"query": "SELECT * FROM information_schema.constraint_column_usage"
}
```

But it returns **no rows at all** (empty result), even though the underlying Postgres database definitely has constraints.

### 3) Importing/using the server library (introspection during cube build)

I also tried importing the server library and noticed the failure seems related to **accessing certain system tables/views** (like the example above). When those queries fail or return empty, cubes cannot be built correctly / the introspection-based parts break.

## Environment

* Database: **PostgreSQL**
* Deployment: Docker / Node
* Works the same **with existing Data Models** and **without any Data Models** (problem persists either way)

## Expected behavior

* A supported/reliable way to fetch DB schema metadata from Cube for Postgres, including:

* tables and columns
* **primary keys**
* **foreign keys / relationships**
* `information_schema` queries should return expected rows when executed via CubeSQL (or clear documentation if they are intentionally blocked).
* If `playground/db-schema` is not intended to be used, guidance on the correct endpoint / recommended approach.

## Actual behavior

* `playground/db-schema` returns incomplete schema data (PK/FK missing or incorrect).
* `SELECT * FROM information_schema.constraint_column_usage` via `/cubejs-api/v1/cubesql` returns no rows.
* Introspection seems to fail on certain metadata tables/views, preventing correct cube/data model generation.

## Questions

1. Is this a known limitation/bug for Postgres schema introspection in Cube?
2. Are `information_schema` / `pg_catalog` queries restricted/filtered when running through CubeSQL? If so, is there a way to allow them?
3. What is the recommended/supported way to query a database schema through Cube?
4. Is there an official way to **create Data Models programmatically** (or generate them from schema metadata) that Cube supports?

## Reproduction steps (high level)

1. Connect Cube to a Postgres DB with PK/FK constraints.
2. Call `playground/db-schema` and check whether PK/FK are present/correct.
3. Call `/cubejs-api/v1/cubesql` with:

```sql
SELECT * FROM information_schema.constraint_column_usage;
```
4. Observe empty results and incomplete/failed introspection.

## Additional info / logs

If needed, I can provide:

* Cube logs around schema introspection / cube build
* Postgres version and sample schema (sanitized)
* The exact Cube config and connection settings