Azure Agentic retrieval fails with unhelpful error response in spite of correct setup

When using the [example python notebook provided for agentic search](https://github.com/Azure-Samples/azure-search-python-samples/tree/main/Quickstart-Agentic-Retrieval), after having configured everything correctly, I get the following response:

```
HttpResponseError Traceback (most recent call last)
[](https://localhost:8080/#) in ()
12 })
13
---> 14 retrieval_result = agent_client.retrieve(
15 retrieval_request=KnowledgeAgentRetrievalRequest(
16 messages=[KnowledgeAgentMessage(role=msg["role"], content=[KnowledgeAgentMessageTextContent(text=msg["content"])]) for msg in messages if msg["role"] != "system"],

3 frames
[/usr/local/lib/python3.11/dist-packages/azure/search/documents/agent/_generated/operations/_knowledge_retrieval_operations.py](https://localhost:8080/#) in retrieve(self, retrieval_request, x_ms_query_source_authorization, request_options, **kwargs)
219 map_error(status_code=response.status_code, response=response, error_map=error_map)
220 error = self._deserialize.failsafe_deserialize(_models.ErrorResponse, pipeline_response)
--> 221 raise HttpResponseError(response=response, model=error)
222
223 deserialized = self._deserialize("KnowledgeAgentRetrievalResponse", pipeline_response.http_response)

HttpResponseError: () An error has occurred.
Code:
Message: An error has occurred.
```

### Steps I followed:
- Clone the notebook
- Created the AI search index on the "Standard" tier
- Created an Open AI service
- Created the gpt-4o deployment
- Used `text-embedding-3-large` embedding
- Added the following permissions for the user - Search Service Contributor, Search Index Data Contributor, Search Index Data Reader
- Added all the necessary keys
- Executed your notebook code

Empieza volviendo a ejecutar el notebook Quickstart-Agentic-Retrieval mediante la llamada agent_client.retrieve, utilizando los detalles de configuración indicados en el issue: Search tier, OpenAI deployment, embedding model, permissions y keys. Compara la solicitud que falla con la configuración del notebook. Se considera terminado cuando el ejemplo completa la recuperación o proporciona un error específico y procesable en lugar del HttpResponseError genérico.

Escrito por el modelo de indexación a partir del texto del issue.