[serve] Documentation on how to use model registries/feature stores

[edoakes]

One big advantage of Serve being python-native is easy integration with model registries (e.g., MLFlow) and feature stores (e.g., Feast). We should call this out in a separate docs page.

[jinnovation]

feature stores (e.g., Feast)

Big +1 to this. IMO this is a pretty noticeable gap in the Ray Serve documentation and overlooks one of the biggest friction points when it comes to model serving—feature access.

Demonstrating use of Ray Serve in tandem with select model registry + feature store solutions, e.g. in the Examples section, would go a long way towards demonstrating Ray Serve's ability to integrate will with broader platform offerings—or maybe surface shortcomings to address. 😅