Add runs_on input to reusable deploy workflows

[Christdej]

Summary

Adds an optional runs_on input to the reusable deploy workflows so consumers can pick a self-hosted runner (or any label other than ubuntu-latest) for the docker build / image copy jobs.

Affected workflows:

• build_and_publish_docker_image_to_container_registry.yml — build job.
• deploy_to_development.yml — plumbs runs_on down to the publish call.
• deploy_to_staging.yml — plumbs runs_on down to the publish call.
• promote_to_production.yml — image-copy job.

Default remains ubuntu-latest. Backward compatible.

Motivation

sara-anonymizer builds a ~10 GB image (torch + torchvision + deeplab weights) that runs out of disk on GitHub-hosted ubuntu-latest runners (14 GB total). Its pre-migration workflow used runs-on: ubuntu — an Equinor self-hosted runner label with more disk. Since #91 migrated the publish step into the armada reusable, sara-anonymizer's dev deploys have been failing with:

System.IO.IOException: No space left on device

With this input, sara-anonymizer (and any other repo with the same issue) can pass runs_on: ubuntu in its deploy workflow to route the build back to the self-hosted runner.

Notes

• The kustomize update job (update_kubernetes_deployment.yml) still runs on ubuntu-latest — it does a lightweight git checkout/sed/push and never needs the extra disk.
• Follow-up PR will update sara-anonymizer to set runs_on: ubuntu.