[AI Compose] Support many model deployments for AI Service

[vhvb1989]

Current approach is to use one AI Service per model. It simplifies the e2e experience when folks don't need to think about connections of services and just list the models to deploy.

For this issue, we want to support the default case of just listing the models in azure.yaml as a way to define all models to the same connection/service.

And then, introduce the schema for modeling more than one connection if folks want to do that.