Use transforming informer for cluster cache #19199
Description
Summary
Today we use custom list/watch logic in the cluster cache component of gitops-engine to monitor resources. We did that originally to avoid storing a bunch of state in memory that we don't actually need (we just need resource metadata, not contents, in the cluster cache).
0.29 Kubernetes libraries introduced a transforming informer which could allow us to implement the same logic but using standardized tooling.
Motivation
We get the benefits of using upstream Kubernetes libraries instead of rolling our own. An example of a benefit is randomized watch timeouts. In the current cluster cache, we hardcode watch timeouts at 10min. The informer randomizes the timeout to spread load. We could re-implement the randomization, but it's nice to take advantage of upstream functionality.