Alerting: Expose per-rule full processing duration

### Summary When configuring or reviewing alert rule intervals in Grafana Alerting, it is difficult to determine if a rule is evaluating too frequently compared to the time it takes to fully process. This can lead to dropped ticks, resulting in misleading or unexpected alert behavior. ### Problem If a rule’s evaluation interval is shorter than its full processing duration, we see dropped ticks and potentially confusing behavior, such as: - Alerts expiring incorrectly before they are handled by Alertmanager - Delays in updating state transitions - Inconsistent application of labels/annotations due to templating delays The Last evaluation duration, exposed in the UI, only reflects the query evaluation time, not the entire rule processing duration (including state management, templating, etc). A dropped ticks metric is available (`schedule_rule_evaluations_missed_total`), but it only signals that a problem exists and does not provide timing context to help determine a safer interval. A process duration metric is available (`rule_process_evaluation_duration_seconds`), but the metric aggregates across all rules, making it difficult to identify individual problematic rules. ### Proposed Solutions Some potential options (not exhaustive): 1. Add a rule-level label to `rule_process_evaluation_duration_seconds` metric. However, this is a histogram metric, so this option is fairly heavy cardinality-wise. 2. Add a new gauge metric for rule-level process duration. This approach has much lower cardinality than a histogram while still providing meaningful insight over time. Depending on the scrape interval, some individual evaluations may be missed; however, for the primary use case of detecting rules whose processing time is long relative to their configured evaluation interval, this metric is sufficient and actionable. 3. Add process duration alongside the existing `EvaluationDuration` in rule state. This approach captures only the most recent process duration without keeping …