Grafana fails to start completely when provisioned alert rule references non-existent receiver

### What happened? ## Summary When an alert rule is provisioned (e.g., via ConfigMap) that references a contact point/receiver that doesn't exist, Grafana fails to start entirely. This is overly disruptive behavior — a single misconfigured alert rule should not prevent the entire Grafana instance from starting. ## Actual Behavior Grafana refuses to start with error: ``` level=error msg="Failed to provision alerting" error="alert rules: invalid alert rule\nreceiver MyFancyAlertChannel does not exist" Error: ✗ invalid service state: Failed, expected: Terminated, failure: invalid service state: Failed, expected: Running ``` ## Impact This behavior is particularly problematic in multi-tenant or platform scenarios where: - Teams are allowed to define their own alerts via provisioning (GitOps) - A single team's misconfiguration brings down the shared Grafana instance for everyone - Recovery requires manual intervention to fix or remove the faulty configuration ### What did you expect to happen? ## Expected Behavior Grafana should start successfully and either: - Log a warning/error about the invalid receiver reference - Disable or skip the misconfigured alert rule - Fall back to a default receiver The invalid alert rule should **not** prevent the entire application from starting. ## Suggested Improvement - Implement graceful degradation: start Grafana but mark invalid alert rules as broken/disabled - Provide clear visibility in the UI for misconfigured rules - Consider a "strict mode" flag for environments that prefer fail-fast behavior ### Did this work before? I don't think so. ### How do we reproduce it? ## Steps to Reproduce 1. Create a contact point via the Grafana UI (e.g., `MyFancyAlertChannel`) 2. Provision an alert rule via ConfigMap/file provisioning that references this UI-created contact point 3. Redeploy/restart Grafana (the UI-created contact point is not persisted in provisioning and is lost) 4. Grafana fails to start ### Is the bug inside a dashboard panel…