[batch] Build resource provider layer for batch workloads

[Jeffwan]

🚀 Feature Description and Motivation

Now, we need to launch kubernetes jobs and host GPU models. In order to support broader customers. We need to consider other resource providers. For example, we can launch models in Sagemaker or SkyPilot etc, the resource provider interface should be abstracted.

Use Case

batch api to fit different platforms

Proposed Solution

No response