ci: gate chart version bumps and helm-docs freshness, report on the PR

[kylecrawshaw]

What

Adds a helm job to ci.yaml with two checks, both reported as a sticky comment on the PR:

• Chart version bump (PR-only): for each charts/*/ directory changed relative to the merge-base with the PR base branch, version: in that chart's Chart.yaml must differ from the base. New charts pass automatically.
• helm-docs freshness (every event, so it also gates releases through workflow_call): regenerates docs with helm-docs pinned at v1.14.2 and fails if any charts/*/README.md changes, printing the diff and pointing at make helm-docs.

Findings land in a single marker-identified PR comment that is updated in place on later pushes (no comment stacking) and flips to "checks passed" once fixed. If the job fails for infrastructure reasons with no findings, an existing comment is left untouched. On fork PRs the token is read-only, so the comment degrades to a log warning and the job failure alone carries the news. pull-requests: write is scoped to this job only.

Why the chart changed too

The check surfaced its own flake source on the first local run: the chart README's footer embedded the generating helm-docs version (Autogenerated ... helm-docs v1.14.2), so any differently-built helm-docs (e.g. a go install build with no version metadata) regenerated a spurious footer diff. This drops the versionFooter template from README.md.gotmpl so output is deterministic across tool builds, regenerates the README, and bumps the chart to 0.2.1 — as the new check itself requires for any chart directory change.

Testing

• Version check exercised locally against origin/main both ways: passes with the 0.2.1 bump, correctly flags the chart when the version is reverted to 0.2.0.
• helm-docs regeneration leaves a clean tree after the footer fix.
• Workflow YAML parses. This PR's own CI run is the end-to-end test of the job, including that the pinned v1.14.2 binary renders byte-identical output.

🤖 Generated with Claude Code

[github-actions]

Helm chart checks passed

Earlier findings on this PR are resolved.

[anders94]

LGTM