Split Config-Server into two pieces

[faebr]

Currently config-server and data-server run in the same pod. If one crashes, the whole pod will crash and restart and stuck in a loop. During this time the api service which stores the config is also unavailable, making it difficult to debug which config made the crash or to edit the config. Thus we should split front-end and back-end as discussed during the community call on October 27th.

[alexandernorth]

I believe we should end up with 3 pods under this approach: data-server, config-server, api-service (naming tbd)

[henderiw]

I would do it in a 2nd step, as we need to scale out the data server for big deployments and it open a bunch of extra design considerations when we split DS/CS controller