Replace the current autoscaler with Ray implemetation

[asitang]

Ray recently included a ray Serve autoscaler in their offering, which scales up and down the replicas based on the network traffic. We have a basic version of it implemented in-house (https://github.com/EDRN/labcas-ml-serve/tree/main/deployments/autoscaler). We need to replace it with the Ray native version.
ref: https://docs.ray.io/en/latest/serve/architecture.html