Add Kubernetes models Recipe - Bicep

Overview of the Resource type and Recipe

Radius.AI/models only has an Azure recipe today (Azure OpenAI, via the AKS recipe pack). Create a Kubernetes Recipe in Bicep using an OpenAI-compatible in-cluster inference server.

Acceptance criteria

1. A Kubernetes Bicep recipe at AI/models/recipes/kubernetes/bicep/ (e.g. LocalAI/Ollama: Deployment + Service) serving an open-weight model mapped from model.
2. Emits the Radius result object with the resource type's outputs.
3. Recipe entry added to the default Kubernetes recipe pack (recipepack/kubernetes/default-recipepack.bicep) and wired into the publish-bicep-recipes workflow matrix.
4. README; manual testing with make test-recipe.

Additional context

Footprint: default to a small CPU model for testing; model pull at startup can be slow. Azure-style model values (e.g. gpt-5-mini) don't map 1:1 to open weights — pick an open-weight default. Example: Data/postgreSqlDatabases/recipes/kubernetes/bicep/; contribution guide docs/contributing/contributing-resource-types-recipes.md.