Is your feature request related to a problem? Please describe.

1. Confusion for new users: Most kubernetes users already have a cluster and may already have GPU nodes available. Kaito’s current workflow is optimized for the case where GPU lifecycle is bound to inference server lifecycle. It's an anti pattern for k8s. e.g. to deploy an nginx pod, typically k8s users do not expect k8s to create a new node for the pod to run.
2. Time takes to deploy a new model: To deploy a new model, the current GPU provisioner experience takes much longer than bring your own node experience since it also includes node provision time.

Describe the solution you'd like

Describe alternatives you've considered

Additional context