The controller was never explicitely enabled, although it must have
been working previously, possibly enabled silently inside vanilla code.
It stopped working when we released 1.31, which in turn disabled
automatic rescheduling of pods from unavailable nodes.

feat: be able to give several anchor

See merge request !21

The garbage collector was failing to handle CRDs due to not being
able to create an informer for CRD types, which in turn was breaking
garbage collection altogether.

The issue was related to using the wrong informer factory (a core
factory from apimachinery instead of the dynamic factory from the
controller-manager module)

This reverts commit 12312f38.

Previously the plugins initialization was done for the apiserver
config only, which enabled admission for core resources handled by
the apiserver, but the extensions server lacked the admission chain
and did not apply webhooks to CRDs