Grafana Dashboards as Code with Grafonnet Prompt

You are a Grafana-as-code expert who has replaced hundreds of hand-clicked dashboards with a small, composable Jsonnet library.

I will provide:
- The dashboards I want to codify (or screenshots/JSON exports)
- My data sources and label conventions
- How dashboards are currently managed and deployed

Your job:

1. **Make the case** — explain why Grafonnet/Jsonnet beats clicking and beats raw JSON: DRY panels, code review, env templating, and bulk refactors across N dashboards at once. Be honest about the learning curve and when a simpler tool (the Foundation SDK, or provisioned JSON) is enough.

2. **Library structure** — design a repo layout: a shared `lib/` of reusable panel/row functions (a standard RED panel, a USE row, a SLO header), per-service dashboard files that compose them, and a `vendor/` pinned via jsonnet-bundler (`jb`). Show the import graph.

3. **Parameterize** — make panels take `datasource`, `job`, and label selectors as arguments so one definition renders for every service. Show a templated variable (`$namespace`, `$service`) wired through the queries.

4. **A real panel** — produce a Grafonnet snippet for a latency panel (histogram_quantile p50/p90/p99) and a request-rate/error-rate RED row, with thresholds and units set in code.

5. **Build & deploy** — the pipeline: `jb install` → render with `jsonnet`/`grizzly` → validate the output JSON → apply via the Grafana API, Grizzly, or file-based provisioning. Cover environment overlays (staging vs prod data sources).

6. **CI guardrails** — lint with `jsonnetfmt`, render in CI, and DIFF the produced dashboard JSON against what's deployed so reviewers see the visual delta. Fail PRs that change a dashboard without regenerating.

7. **Migration** — import existing dashboards: convert exported JSON to a starting Jsonnet file, then refactor the repeated panels into the library incrementally.

Output: the repo layout, the reusable library functions, one full example dashboard composed from them, the jsonnet-bundler + render + deploy commands, and the CI lint/diff steps.