feat(intergral/alerting): HA partitioning

[jhawksley-intergral]

Summary

Adds HA (high-availability) partitioning for the unified alerting scheduler so that alert rule evaluation is distributed across cluster members using consistent hashing.

• Rule partitioning: Each instance evaluates only its share of rules, determined by FNV-64a(orgID:ruleUID) % clusterSize. Partitioning is gated on a minimum cluster size and automatically falls back to full evaluation if the cluster is unhealthy.
• Remote state sync: A background loop periodically loads alert state from the database for rules owned by other instances, so the API can return correct LastEvaluation / rule status regardless of which instance serves the request.
• Topology-aware re-fetch: The fetcher detects cluster membership changes and forces an immediate rule re-fetch, redistributing rules without waiting for the next tick.
• Configuration: New [unified_alerting.ha_scheduler] section in defaults.ini with enabled, min_cluster_size, and remote_state_sync_interval options.

Files changed

Area	Files	What
Core	schedule/partitioner.go	FNV-based rule partitioner with memberlist/Redis peer support
Scheduler	schedule/schedule.go, schedule/fetcher.go	Wire partitioner into fetch loop, topology-change detection
State	state/manager.go	Remote state sync loop, RuleFilter interface, stale cleanup
API	api/prometheus/api_prometheus.go	Fall back to cached state for remote rule status
Config	conf/defaults.ini, setting/setting_unified_alerting.go	HA scheduler settings
Wiring	ngalert.go, notifier/multiorg_alertmanager.go	Plumb partitioner through DI
Tests	partitioner_test.go, fetcher_test.go, manager_remote_state_test.go	Unit + integration tests for partitioning, topology changes, remote sync
Other	expr/mathexp/parse/lex_test.go	Fix unrelated flaky race condition in lexer test
CI	docker_build.yml	Sanitize branch names for Docker tags

Test plan

• Unit tests for partitioner: deterministic assignment, position-based filtering, unhealthy cluster fallback
• Topology-change tests: member join/leave, healthy↔unhealthy transitions, rapid membership changes
• Fetcher integration tests: topology change triggers re-fetch, stable topology skips
• Remote state sync tests: periodic refresh, stale state cleanup, context cancellation
• Manual verification: rule status API returns correct data from non-owning instance

🤖 Generated with Claude Code