ci: upgrade the GitHub Actions to versions that run on Node 24

[charliedowler]

What this does

The workflows use action versions from 2021 and 2022. These actions run on the node12 and node16 runtimes inside the runner. GitHub removes old runtimes over time. When that happens the chart release stops and the PR checks stop. Self-hosted users who wait for a chart release notice first.

This PR bumps each action to the current major version. It also removes one dead job. It changes no working behaviour.

The bumps

Workflow	Action	Before	After	Runtime after
all	actions/checkout	v2, v2.2.0	v7	node24
lint-and-test	actions/setup-python	v2, python 3.7	v7, python 3.13	node24
lint-and-test, ci-release	azure/setup-helm	v1	v5	node24
lint-and-test	helm/chart-testing-action	v2.2.0	v2.8.0	composite
lint-and-test	helm/kind-action	v1.2.0	v1.14.0	node24
ci-release	helm/chart-releaser-action	v1.1.0	v1.7.0	composite
generate-helm-docs	stefanzweifel/git-auto-commit-action	v4.13.1	v7.2.0	node24

Every remaining action runs on node24 or runs as a composite action with no Node runtime.

Removing the automerge job

The automerge job never runs, so this PR deletes it and drops pascalgn/automerge-action with it. Its condition tests contains(github.event.pull_request.labels, 'merge after testing'), but labels holds objects and not strings, so the test is always false. The job reports skipped in all of the last 25 workflow runs. The action is also a dead end for this upgrade: v0.16.4 is the newest release, it runs on node20, and the repository has no commit since September 2024. main has no branch protection and no ruleset, so no required check depends on this job. The workflow keeps its current name to hold the check names steady.

What stays the same

• The chart releaser runs on a push to main.
• CR_SKIP_EXISTING: true stays.
• The kind Kubernetes matrix stays at v1.21.2, v1.22.5 and v1.23.3.
• The helm versions stay at v3.4.0 and v3.7.2.
• The ct lint and ct install commands stay.
• No chart version changes. This PR needs no chart release.

Breaking input checks

I compared the action.yml inputs at the old tag and the new tag for each action.

• helm/chart-releaser-action v1.7.0 adds a skip_existing input. The action appends --skip-existing only when that input is not empty. The input stays unset, so cr reads CR_SKIP_EXISTING from the environment as before. The env var keeps working.
• helm/kind-action renames log_level to verbosity. The workflow does not set it. node_image and cluster_name are unchanged.
• azure/setup-helm makes the token input optional. Both steps pin an explicit helm version, so the action makes no API call for the latest release.
• actions/setup-python v7 does not supply python 3.7. Python 3.7 is end of life. The version moves to 3.13. chart-testing-action still needs python because it installs yamllint and yamale with pip.
• stefanzweifel/git-auto-commit-action v7.2.0 keeps every v4 input. The defaults for the commit author change from GitHub Actions <actions@github.com> to github-actions[bot]. The workflow sets neither, so the docs commit gets the bot identity. This is the only visible change in behaviour.

Out of scope

docker://jnorwood/helm-docs:v1.7.0 is a container action. It runs no Node runtime, so this PR leaves it alone.

Coordination with #150

PR #150 is open. It rewrites lint-and-test-charts.yml and ci-release.yml, and it already carries most of these bumps. I used the same version strings as #150 so the overlapping lines match.

After #150 lands, two changes in this PR remain new:

• generate-helm-docs.yml, which ci: test installs and upgrades, gate the auto-release, modernise the matrix #150 does not touch.
• The removal of the automerge job, which ci: test installs and upgrades, gate the auto-release, modernise the matrix #150 keeps.

This PR must rebase onto main after #150 merges, before it leaves draft.

Linear: PROD-10245

[linear-code]

PROD-10245