feat(dashboard): add cache and checkpoint observability

[liunyl]

Context

Companion dashboard delivery for tx_service #561. EloqKV operators need node-level visibility into resident/dirty key populations and checkpoint scheduling, durable progress, and failures.

Changes

Updates both EloqKV overview variants with a collapsed six-panel Cache and Checkpoint Observability row:

• resident and dirty key populations by node;
• dirty-key ratio as a ratio of summed populations;
• checkpoint attempt interval average and p95;
• checkpoint durable-advance interval average and p95;
• cumulative checkpoint failures by node and reason;
• the existing consecutive-failure alert signal as a node-level panel.

The existing is_continuous_checkpoint_failures > 0 alert rule is preserved.

Verification

jq empty <both dashboard files>
  passed
unique panel-ID check
  passed for both files
new-row panel count check
  passed: exactly six panels in each file
PromQL inspection
  passed: node_id aggregation, histogram le grouping, guarded denominators
live Grafana import + three-node Prometheus scrape
  passed

Manual validation confirmed dirty counts rise after writes and return to zero after checkpoint while resident counts remain, and checkpoint attempt/advance counters continue increasing on a synchronized standby.

Dependency and rollout

Requires the metrics added by tx_service #561. Deploy the tx_service change before relying on these panels; older servers will show no data for the new series. Rollback is a revert of this PR.