feat(mcp): query Google Cloud metrics live with PromQL

apps/ai/src/mcp/dispatcher.ts

@@ -2,6 +2,7 @@ import { AuditLogService } from "@maple/backend/services/audit/AuditLogService"
 
 import { WarehouseQueryService } from "@maple/backend/services/warehouse/WarehouseQueryService"
 import { VcsSourceService } from "@maple/backend/services/integrations/vcs/VcsSourceService"
+import { GcpPromQlService } from "@maple/backend/services/integrations/GcpPromQlService"
 import { SandboxClient } from "@maple/backend/sandbox/client"
 import { CloudflareRepoSandboxLive } from "@maple/backend/services/sandbox/CloudflareRepoSandbox"
 import type { SandboxError } from "@yielded/agent/sandbox"
@@ -295,6 +296,7 @@ const McpRuntimeServicesLive = Layer.mergeAll(
 	ErrorIssueWorkflowService.layer,
 	ErrorPolicyService.layer,
 	ErrorsService.layer,
+	GcpPromQlService.layer,
 	IssueFixVerificationService.layer,
 	QueryEngineService.layer,
 	RecommendationIssueService.layer,

apps/ai/src/mcp/resources/instructions.ts

@@ -109,6 +109,7 @@ first occurrences ──> (candidate) ──> issue in \`triage\`
 - Performance analysis: find_slow_traces -> inspect_trace -> get_service_top_operations
 - Trend analysis: query_data (timeseries or breakdown)
 - Service discovery: list_services -> diagnose_service
+- Google Cloud metrics, live (Cloud Monitoring through Google's PromQL API, not the ingested \`gcp.*\` warehouse metrics): find_gcp_metrics search=... -> find_gcp_metrics metric=... (kind, labels, starter query) -> query_gcp_promql. Aggregate with \`sum by\`/\`avg by\`: Google bills per series returned
 - Infrastructure (hosts, Kubernetes pods/nodes/workloads, Docker containers): list_infra (no \`kind\` = overview of what reports, hottest first) -> inspect_infra (CPU/memory over the window with the peak time, a node's or workload's pods). diagnose_service shows the workload and pods a service runs on. Resource metrics are also chartable: create_dashboard group_by \`resource.host.name\` / \`resource.k8s.pod.name\` on source=metrics
 - Deploy verification: service_deployments (versions per service, newest vs previous) -> compare_periods around the rollout. Route usage over up to 90 days: route_usage. Silence check before calling an outage: ingest_freshness. Volume per service: ingest_usage. Database query shapes by volume: db_query_volume
 - Alert management: list_alert_rules / list_alert_destinations -> get_alert_rule -> preview_alert_rule (dry-run before saving; a window with no data is a blind rule, not a healthy one) -> create_alert_rule / update_alert_rule / delete_alert_rule -> list_alert_incidents