Support multi-node distributed inference

[sdesai345]

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

To serve larger language models with billions of parameters, users want to deploy multi-node inference on their Kubernetes cluster using the KAITO out-of-box presets.

Describe the solution you'd like

Multi-node inference on HuggingFace models with limited node pool configuration steps, starting with support for widely-used models with 70B params or less.

[MartinForReal]

related: #873

[chewong]

Marking this as done. We will track https://github.com/kaito-project/kaito/issues/1145 separately.