Open
Description
Currently, every time we run sky serve status --endpoint, it will contact the sky serve controller to get the endpoint. It is not friendly for a bash-based script that calls this CLI multiple times, say curl $(sky serve status --endpoint svc1)/v1/models. We should have a local database serve as a cache for the service endpoint since it will not change while the service is running.
Notice that we still use the remote database on the SkyServe controller as the only source of truth. If we don't find the service endpoint in the local cache, then we contact the SkyServe controller for the endpoint.