Derive liveness cadence only from admissible scheduler groups

Description

Wave 3 found that fastest_effective_interval_seconds includes profile-excluded and shed groups. Under the availability profile, an excluded faster group can make the process fail liveness before its only admissible group is due.

Acceptance Criteria

• #1 The fastest interval considers only enabled, profile-allowed, non-shed groups
• #2 An empty admissible set falls back to the configured resolve interval
• #3 A regression test covers the availability-profile 180-second versus 270-second boundary

Definition of Done

• #1 just check (ruff format –check, ruff check, mypy, generated-doc drift, offline API conformance, and the marker-filtered pytest run with the 80% coverage floor — this is exactly what the CI test job runs)
• #2 just gen, when metrics, config, endpoints, collectors, the settings schema or the chart config changed — just check includes the drift gate and CI fails the build on it
• #3 Grafana queries in grafana/dashboards/*.json and grafana/alerts/ updated, if a metric or label name changed

Implementation Plan

Add the availability-profile liveness regression, filter the fastest cadence to admissible non-shed groups, and run focused scheduler and liveness tests.

Implementation Notes

The failing availability-profile regression observed 60 seconds from an excluded priority-2 group instead of 300 seconds from the admissible priority-1 group. The liveness cadence now uses enabled/profile-allowed/non-shed groups and falls back when none exist. Integrated gate: 2827 passed.

Final Summary

Made liveness cadence profile-aware and shed-aware with a test-first availability-profile regression; just check and just ci pass.