Fix/ci periodic

[zvdy]

Fix periodic CI reliability and prevent resource waste

Problem

Solves #488 and guardrails common failed CI's, some samples here:

• https://github.com/GoogleCloudPlatform/kubectl-ai/actions/runs/16967900774/job/48097655096
• https://github.com/GoogleCloudPlatform/kubectl-ai/actions/runs/17341505682/job/49236025751
• https://github.com/GoogleCloudPlatform/kubectl-ai/actions/runs/16941678334/job/48012177075

1. Checksum validation errors from outdated kind-action version
2. kubectl download timeouts causing cluster setup failures
3. Evaluation timeouts (30+ minutes) causing job failures and overlapping runs

Changes

• Upgraded to latest stable version: helm/kind-action@v1.12.0 releases
• Increased cluster wait timeout from 60s to up to 300s for better reliability
• Optimized Timeouts
  • Job timeout: 12 minutes (fits within 15-minute cron schedule)
  • Cluster setup: 3 minutes timeout
• 2 retry attempts for evaluation step
  • 4-minute timeout per attempt with process cleanup
• 10-second wait between retries

---

TODO

I'm not able to test this on a GKE cluster as the setup is done to replicate, so I would kindly ask you @droot whenever you have time to clone this branch/commits and test the action a few times on workflow dispatch and see if it solves the issues. The checksum issues were due to the reference of the commit itself instead of a release in .github/actions/kind-cluster-setup/action.yaml so we should be good now :-)

[droot]

Thanks @zvdy . Getting the PR merged because they improve our current set-up. There are some other changes that I merged today #529 that should reduce the resource requirements. So I am hoping with these, flakes should be reduced.