`JupyterLabJobService` endpoint returns “User <UPN> does not have access to node <node-id>” on running AML job

- **Package Name**:
- azure-ai-ml
- **Package Version**:
- 1.25.0
- **Operating System**:
- Windows 11 Enterprise (24H2) x86_64
- **Python Version**:
- 3.12.9

## Describe the bug
When enabling `JupyterLabJobService` for an Azure ML v2 `command` job running on a **compute cluster** (not a compute instance), the job runs successfully and the JupyterLab service endpoint is returned by `ml_client.jobs.show_services(...)` / `az ml job show-services`. However, opening the returned JupyterLab URL in a browser shows an authorization page stating that the user does not have access to the node.

This prevents using `JupyterLabJobService` for interactive debugging (terminal/REPL) on cluster jobs.

## To Reproduce

1. Create an Azure ML v2 `command` job targeting a compute cluster.
2. Enable `JupyterLabJobService` on the job, e.g.:

```{python}
from azure.ai.ml import command
from azure.ai.ml.entities import JupyterLabJobService

job = command(
code="./src",
command="python train.py; sleep infinity",
environment="",
compute="",
services={"jupyter": JupyterLabJobService()},
)
```

3. Submit the job and wait until it reaches `Running`.
4. Retrieve the service endpoint:

* Python: `ml_client.jobs.show_services("", node_index=0)`
* or CLI: `az ml job show-services --name --node-index 0`
* or use the AzureML UI

5. Open the returned JupyterLab URL in a browser or the hyperlink in the UI.

## Expected behavior

The JupyterLab page loads (after authentication) and provides access to a terminal/REPL inside the running job container.

## Actual behavior

Opening the returned JupyterLab endpoint shows an error page indicating that the signed-in user does not have access to the node.

## Screenshots

## Additional context

* The job itself runs successfully; only the interactive endpoint is inaccessible.
* Region: West Europe.
* Compute runs on a VNet

Inizia riproducendo il problema con azure-ai-ml 1.25.0 usando un command job su un compute cluster e JupyterLabJobService, quindi traccia i percorsi JupyterLabJobService e ml_client.jobs.show_services. Confronta il comportamento tra cluster e compute-instance; il lavoro è completato quando l’endpoint restituito apre una sessione JupyterLab autenticata per il container del job in esecuzione senza l’errore di accesso al nodo.

Scritto dal modello di indicizzazione a partire dal testo della issue.