Run user submission using k8s

[pauloguilhermepp]

Run User Submissions Using Kubernetes

This PR aims to update Codabench to run user submissions using Kubernetes. The default way to run submissions is still using Docker containers.

This PR is still in progress, and we would appreciate your feedback.

Main Changes

To run submissions using Kubernetes, we mainly:

• Added Kubernetes Python client to dependencies.
• Updated compute_worker.py to create and delete Kubernetes pods.
• Updated permissions granted to the compute-worker pods (rbac files).

Issues this PR resolves

• Closes Running User Submissions on Kubernetes #2152

How to Test it

Deploy Codabench using Kubernetes as explained at #2055.

Observations

• This PR is a draft and should be merged after Deploy codabench using kubernetes #2055. It's already open just so that we can get your feedback.
• This branch was created from the branch used in Deploy codabench using kubernetes #2055. After closing that PR, this branch will be rebased by pulling commits from develop and tested again.
• Inside of compute_worker.py, CONTAINER_ENGINE defines who is running the submissions (docker, podman, or kubernetes). Its default value is "docker". The changes applied here should only affect the compute-worker if CONTAINER_ENGINE="kubernetes".

Checklist

• Code review by me
• Hand tested by me
• I'm proud of my work
• Code review by reviewer
• Hand tested by reviewer
• CircleCi tests are passing
• Ready to merge

[Didayolo]

As mentioned in #2268:

• This PR needs to be rebased. It uses the old container files names, and there is many conflicts to be resolved. Please do not merge this one before Compute worker - Improve status update and logs #2223 so we resolve the conflicts from the latest version of the compute worker.

• The change in the main Containerfile needs to be adjusted, as the build is not working consistently. It is probably easy to fix as it is just package install failures. Generally it is recommended to pin the base image (node:lts-alpine3.23 is not exactly pinned because of node:lts), and to use npm ci instead of npm install.

[pyxelr]

Thanks for this work @pauloguilhermepp! We deployed Codabench on an enterprise EKS cluster where Docker-in-Docker is blocked by Kyverno security policies (no privileged containers). This PR was the perfect solution - full E2E submission execution working with the K8s-native compute worker, no policy exceptions needed. Looking forward to this getting merged!