Fix FetchModelDefinitionsService to handle SM

What does this MR do and why?

Fix GraphQL crash in GitLab Duo Self-Hosted model configuration after v18.4 upgrade

Issue: After upgrading to GitLab v18.4, customers lose complete access to Duo Self-Hosted model configuration due to GraphQL errors. The FetchModelDefinitionsService returns nil instead of a ServiceResponse object, causing NoMethodError: undefined method 'payload' for nil:NilClass in the GraphQL resolver.

Root Cause: The service checks the wrong feature flag (:ai_model_switching for group-level selection) when self-managed instances need instance-level model definitions regardless of group-level settings.

Fix:

1. Remove :ai_model_switching check from FetchModelDefinitionsService - only check duo_features_enabled
2. Always return ServiceResponse objects (never nil) with graceful error handling
3. Update GraphQL resolver to safely handle ServiceResponse objects

Impact:

• Fixes complete Duo feature outage for self-managed customers
• Allows instance-level model selection without requiring group-level features
• Provides graceful degradation for network/API errors

Related Issues:

• Closes #571212 (closed)
• Customer tickets: https://gitlab.com/gitlab-com/request-for-help/-/issues/3444

How to set up and validate locally

• Try to edit or create a self-hosted model, with ai_model_switching FF disabled.

MR acceptance checklist

Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.