Introduce Repository Layer to `Agent` domain GQL resolvers

## Summary

Currently, the repository layer is only partially applied to GQL objects related to the Agent domain, or it is not applied at all. The current structure where the API Layer accesses the database directly makes code changes difficult and hinders code reuse. Especially at this point, where we aim to migrate the GQL library from Graphene to Strawberry, introducing layers with distinct roles and responsibilities while minimizing dependencies between them will become increasingly important.

Therefore, we plan to introduce the Repository Layer pattern into the resolvers for GQL objects related to the Agent domain (`Agent`, `AgentNode`, `AgentSummary`). Currently, the Repository Layer in [http://Backend.AI](http://Backend.AI) holds dependencies on multiple Stateful Sources and Cache Sources, and performs the role of combining these sources. Through this refactoring, we aim to separate actions such as accessing the DB to persist data or retrieve persisted data, and actions like storing or retrieving data via the valkey client (whether cached or arbitrary) from the API Layer.

##

JIRA Issue: BA-2725