Add create-cronjob benchmark task to k8s-bench

[wussh]

Description

This pull request adds a new benchmark task to k8s-bench that tests kubectl-ai's ability to create a Kubernetes CronJob resource. This expands test coverage to include more Kubernetes resource types.

Changes

• Created a new benchmark task in k8s-bench/tasks/create-cronjob/
• Implemented task.yaml with a prompt to create a CronJob that runs at midnight
• Added setup.sh, cleanup.sh, and verify.sh scripts for the benchmark
• Designed verify.sh to be robust and handle different valid implementations

Benefits

• Tests kubectl-ai's ability to work with CronJob resources
• Validates understanding of cron schedule syntax
• Adds a medium-difficulty benchmark to the existing suite
• Follows the best practices outlined in the k8s-bench contribution guide

Testing

The benchmark has been prepared according to the format of other benchmark tasks and includes comprehensive verification logic.

Checklist

• Task follows the structure of existing benchmark tasks
• Task includes proper setup, cleanup, and verification scripts
• Verification script handles multiple valid implementation approaches
• Scripts are executable

[noahlwest]

Can we adjust the namespace to just create-cronjob? Including test, eval, etc. can affect how the model handles the request.

[noahlwest]

minor nitpick: Is it reasonable to do a single kubectl get call, save the output, and check against that for the verification checks? The intent you have here is clear, so if that optimization would make things less readable or too tricky, feel free to disregard this comment.

[droot]

Since we are moving k8s-bench to its own repo, closing this out.