I've been digging through our API code in order to automatically document the API, and it's hard to understand.

The division between master and apiserver isn't clean, and the API machinery is spread out between (at least) api, runtime, conversion, version, and registry, with a seemingly arbitrary decomposition into files and a bewildering proliferation of interfaces (storage, codec, scheme, ...). Additionally, it's not obvious which code is used by the server, by clients, or both. Also, ideally, the generic machinery would be cleanly separated from Kubernetes's specific API object implementations.

After we get some API plugins working, I suggest we look hard at how to simplify and reorganize this.

/cc @lavalamp @smarterclayton