JSON Schema source: uri_replace_pattern not applied to nested $ref resolution

[shonigbaum]

Background
Our schema API requires OAuth 2.0 authentication, which the DataHub CLI json-schema source doesn't support natively. We're trying to work around this by using uri_replace_pattern to redirect schema references through an internal authenticating proxy that handles the OAuth flow. However, the feature seems to be broken for nested $ref resolution, making it impossible to ingest schemas that reference other schemas.

Describe the bug
The uri_replace_pattern configuration in the JSON Schema source does not work for nested $ref resolution. While the custom loader is passed to _load_json_schema(), the title_swapping_callback that handles nested references uses its own loader (self.loader) which ignores the configured uri_replace_pattern, causing authentication failures when schemas reference other schemas via HTTPS URLs.

To Reproduce
Steps to reproduce the behavior:

1. Create a JSON schema with a $ref to an external HTTPS URL that requires authentication:

{
 "$id": "https://schemas.example.com/types/person/1.0",
 "properties": {
   "status": {
     "$ref": "https://schemas.example.com/enums/status/1.0"
   }
 }
}

2. Configure the JSON Schema source with uri_replace_pattern to proxy the requests:

{
 "source": {
   "type": "json-schema",
   "config": {
     "path": "/path/to/schemas",
     "platform": "myplatform",
     "uri_replace_pattern": {
       "match": "https://schemas.example.com/",
       "replace": "https://internal-proxy.local/schema?uri=https://schemas.example.com/"
     }
   }
 }
}

3. Run ingestion: datahub ingest run -c config.json

4. Observe error: HTTPError: 401 Client Error: Unauthorized for url: https://schemas.example.com/enums/status/1.0

Expected behavior
The uri_replace_pattern should apply to all $ref resolutions, including nested references. The URL should be transformed to https://internal-proxy.local/schema?uri=https://schemas.example.com/enums/status/1.0 before making the HTTP request.

Root Cause
In json_schema.py (https://github.com/datahub-project/datahub/blob/master/metadata-ingestion/src/datahub/ingestion/source/schema/json_schema.py) line 252-257, the code patches jsonref.JsonRef.callback with title_swapping_callback:

with unittest.mock.patch("jsonref.JsonRef.callback", title_swapping_callback):
   (schema_dict, schema_string) = self._load_json_schema(
       os.path.join(root_dir, file_name),
       loader=ref_loader,
       use_id_as_base_uri=self.config.use_id_as_base_uri,
   )

However, title_swapping_callback (in json_ref_patch.py line 16) calls self.loader(uri) directly, which uses the default jsonref.jsonloader instead of the custom stringreplaceloader configured via uri_replace_pattern.

Environment:

• DataHub version: 1.4.0
• Source: json-schema
• Deployment: Kubernetes

Additional context
The stringreplaceloader is correctly created and passed as ref_loader to _load_json_schema(), but the callback mechanism bypasses it. A fix would need to ensure title_swapping_callback uses the configured loader instead of the default one.