fix: Add logical replication error details to chart alerts

[philippemnoel]

What

Add separate five-minute apply_errors and sync_errors annotations to both logical-replication error alerts, plus a logs_url pointing to the existing subscriber-log diagnosis. This brings the chart's alert details in line with paradedb/mcc#317.

Why

The combined count does not tell operators whether errors occurred during normal replication or initial table synchronization. Notification templates can now display both counts for each affected subscription.

How

Scope annotation queries to the alert's namespace, job, subscription, and the chart's cluster-pod selector. Keep existing expressions, thresholds, durations, and labels unchanged. Document per-alert rendering for grouped notifications and the per-pod maximum/failover semantics of the counts. The OnCall Slack template remains MCC-specific.

Chart versions are generated by the publishing workflow, so no static version bump is needed.

Tests

• helm lint with monitoring and PrometheusRule enabled passed.
• Ran the full prometheus-rule-rendering script locally, including existing rendering checks and exclusion of both error rules.
• Added and passed vmalert-tool v1.148.0 tests against Helm-rendered rules: mixed apply/sync errors, sync-only errors, healthy subscriptions, and isolation by subscription, job, namespace, and cluster pods. Wired the test tool into the existing rendering CI job.
• git diff --check passed.

The Kubernetes integration suite remains for CI. No deployment or notification configuration was changed.