Introduce Backend.AI standard pattern in Agent GQL nodes operation

## Motivation

The current Agent GQL nodes operation implementation in `src/ai/backend/manager/models/gql_models/agent.py` directly mixes GraphQL handlers with database operations, using legacy patterns like `simple_db_mutate` and `simple_db_mutate_returning_item`. This creates several problems:

1. Lack of separation of concerns: Business logic is tightly coupled with GraphQL layer and database operations
1. Inconsistent architecture: Other components (e.g., group, domain) have already adopted the modern
action-processor pattern with service and repository layers
1. Poor testability: Direct database operations in mutations make unit testing difficult
1. Limited reusability: Business logic cannot be easily reused in different contexts (e.g., CLI, internal services)
1. Maintenance challenges: Tight coupling makes refactoring and feature additions risky

Applying this pattern to scaling_group will improve code quality, maintainability, and architectural consistency across the codebase.

## Objective

Refactor the scaling group implementation to follow [http://Backend.AI](http://Backend.AI) 's standard action-processor architecture pattern, ensuring:

1. Complete separation of concerns across layers (GraphQL handlers → Actions → Services → Repositories → Database)
1. Type-safe DTOs for all data transfers
1. Testable business logic with proper unit and integration tests
1. Architectural consistency with other components (group, domain)
1. Zero breaking changes to the GraphQL API
1. All existing functionality preserved and tested

JIRA Issue: BA-3558