metrics: improve Resource Control RU granularity and instance-level client visibility

## Enhancement

The Resource Control Grafana dashboards use `rate(...[1m])` for RU consumption, which smooths short-lived workload peaks and valleys. Their Client sections also lack an instance-level view of current request RU consumption.

Improve both the standard and KeyspaceName dashboard variants:

- Use `irate(...[30s])` for RU consumption graphs and avoid the default 2x query interval multiplier to expose short-lived changes.
- Add a Client RU panel based on `resource_manager_client_request_ru_total`, aggregated by TiDB instance and honoring the existing cluster, instance, and resource-group selectors.
- Calculate net request RU as consumption minus refunds. The metric stores both directions as positive counters; instances without a refund series must remain visible.
- Preserve existing panel IDs, TP/AP visibility, and per-query average semantics, and regenerate dashboard JSON from Jsonnet sources.

The 30-second lookback requires at least two scrapes within that window. Client request accounting is distinct from the server-reported RU totals and can temporarily be negative when earlier consumption is refunded.

Start with the Jsonnet sources for the standard and KeyspaceName Resource Control Grafana dashboards, then inspect their RU graphs and Client sections. Verify the query interval, instance aggregation, selectors, refund handling, panel IDs, TP/AP visibility, and per-query averages; regenerate the dashboard JSON and confirm both variants retain the requested behavior.

Written by the indexing model from the issue text.