Do not require ingress-controller for Gardener setup.

[Gerrit91]

Description

Follow-up of #299.

• The kind-cloud-controller-manager did not fully work as it was not possible to run readiness probes against the spawned envoy proxy containers and therefore it did not patch the service status. Running it in the host network resolved this problem.
• A mutating webhook was added to allow static IPs for the services' type load balancer created by Gardener, dynamically retrieving the correct IPs with .nip.io is really hard to do.
• All mini-lab status patching workarounds are now gone.
• Now we also see issues with Netpols for minio as we did for other extensions in the past, too. Here is the PR to deploy them automatically:
  • Allow accessing S3 service inside the Garden Cluster. gardener-extension-backup-s3#31
  • Allow configuring additional netpol for backup-s3 extension. metal-roles#721
• Here is the cleanup for the mini-lab workaround to create the Virtual Garden kubeconfig: Update actions + remove mini-lab workaround to detect loadbalancer port. ansible-common#51
• kind had to be updated on the runners in order to make this PR work, the mini-lab now pins the kind version.

Used AI-Tools ✨

• deepseek-v4-flash-0731 for the mutating webhook

[Gerrit91]

/cc @l0wl3vel

[l0wl3vel]

Draft PR containing my suggested improvements: #327

Yeah, the workarounds for kind-ccm suck. Realistically we should just switch to metallb/kubevip and use APR advertisements directly into the mini-lab_internal network to create load balancers. I think we even did that for kamaji-lab.

kind-ccm which was kind of difficult to integrate in the first place, so metallb is worth a shot imo.

The custom Go webhook was wild either way😅

[Gerrit91]

Cool. Thanks for taking look. I think I am fine with the proposals and would like to merge them into this PR?

[l0wl3vel]

Sure think, you are welcome. Cherry picked them over. Please squash merge the PR to prevent dragging in the 40MB removed webhook binary

[l0wl3vel]

lgtm