cr listing is inefficient due to v1beta1 objectmeta normalization" issue

(where it was 33% of all apiserver usage)

ref: https://github.com/kubernetes/kubernetes/blob/master/staging/src/k8s.io/apiextensions-apiserver/pkg/apiserver/schema/objectmeta/coerce.go#L33

It's from the time where we didn't have pruning for ObjectMeta for CRDs. so in etcd you can basically have garbage in there. we could enforce a one-time cleanup (migration), then annotate the CRD to skip this insane loop. 

Or we can check whether maybe the old code was never live in a 4.x cluster. Not sure when we added pruning releative to 4.1 release.

Slack thread: https://coreos.slack.com/archives/CC3CZCQHM/p1617973777462900