observability: panel-by-panel review of the Grafana dashboard for multi-server/multi-cluster setups

Using the multi-server/multi-cluster observability sample (`operator/config/samples/opentelemetry-multi/`), review the Klio Grafana dashboard panel by panel.
Until now it's only been validated against a single server/cluster, where nothing is ambiguous.

For each panel, check:

- **Aggregation**: sum/max/avg/rate/increase don't silently fold multiple clusters or servers into one misleading number.
- **Legends**: every series clearly identifies cluster, server, tier, stage, path, percentile, etc. No bare, ambiguous, or duplicate labels.
- **Usefulness**: the metric answers a real operational question. Rework or remove panels that don't; add missing ones where
attributable.
- **Visualization**: right chart type/unit, stays readable (e.g. time series over raw stat when history matters).

Fix any broken or misleading panel. Regenerate dashboard JSON with `task grafana:gen` after changes.

Start with the multi-server/multi-cluster sample in operator/config/samples/opentelemetry-multi/ and inspect the Klio Grafana dashboard panel by panel. Check aggregation, legends, operational usefulness, visualization, and readability across multiple clusters and servers. Update broken or misleading panels, then run task grafana:gen; done means the generated dashboard is unambiguous and useful for multi-server/multi-cluster setups.

Written by the indexing model from the issue text.