Schema pull at runtime

It seems that we have to support a fairly big matrix of `etcd versions x kubernetes versions x(+- crds installed)`.

Looking into autogen in #2 and James's effort in #32, which is a great start into getting this project updated - is this a viable path forward?

All the API descriptors are either available from Github or directly from the apiserver and we _could_ also pull them on demand from there, given the currently installed etcd and k8s version.

Another important addition to k8s since the project start are CRDs, how would we get their schema if not from apiserver directly?