feat: Split +Inf overflow alerts by label-set

[vikin91]

+Inf overflow alerts were hard to act on because they only reported the base histogram metric, not the specific label combinations causing the overflow. In practice, one problematic series was hidden among many label variants, so operators could not quickly identify the offending path.

For example, the issue in:

rox_central_k8s_event_processing_duration_count{Action="SYNC_RESOURCE",Dispatcher="unstructured.Unstructured",Resource="ComplianceOperatorRule"}

was being reported as an issue with

rox_central_k8s_event_processing_duration_count

however, the metric file contained over 1k lines for the metric rox_central_k8s_event_processing_duration_count and the particular set of labels was difficult to find.

We fixed this by evaluating histogram series per label-set and changing reporting behavior:

• RED/YELLOW: emit one alert per problematic label-set, with labels included in the title:
  • <metric>{...labels...} (+Inf overflow check)
• GREEN: keep output concise by emitting a single metric-level summary when all label-sets are healthy.
• Add and render Prometheus HELP as Metric description before Message in markdown/console/TUI.

This makes overflow alerts actionable (exact offending series is visible) without adding noise when metrics are healthy.

Moreover, we will now try to guess the histogram unit based on the metric name and the help text.