feat(gastown): measure true container cold-start and mayor-ready latency

The prior container-startup panels measured DO→container RPC round-trips
(/health and /agents/start), not actual cold-start time — /health is
truncated at a 5s client timeout so p99 was bounded below the true
cold-start budget, and the dashboard queries filtered on blob8/blob9
values that don't exist in the AE schema so the panels showed nothing.

This replaces those with two metrics that answer the original questions:

1. container.cold_start — TownContainerDO.warmUp() invokes the
   Container class's startAndWaitForPorts() directly and times it.
   Emitted only when the container was actually started (state !=
   healthy), so the quantiles reflect real cold starts without being
   capped by an arbitrary client-side timeout.

2. mayor.session_ready — container stamps mayorReadyAt when the first
   mayor agent transitions to 'running' and exposes it via /health.
   Town DO reads it and emits durationMs = mayorReadyAt - startedAt
   exactly once per container lifetime (deduped in DO storage keyed
   by containerStartedAt).

Dashboard fixes:
- Rename 'Container Startup Latency' row to 'DO → Container RPC
  Latency' and clarify panel titles so operators don't read p99 off
  those and think it's cold-start time.
- Fix broken success/failure filters: blob8='ok' / blob9='true' →
  blob5='' (error absent), blob5!='' (error present).
- Convert quantile queries from label-column style (which collapsed
  all three series to 'latency_ms') to column-name style (AS p50/p90/
  p99), so the legend actually distinguishes the percentiles.
- Add new 'Container Cold Start & Mayor Ready' row with p50/p90/p99
  panels for the two new events.

Author: jrf0110

services/gastown/container/src/control-server.ts

@@ -10,6 +10,7 @@ import {
   activeServerCount,
   getUptime,
   getStartTime,
+  getMayorReadyAt,
   stopAll,
   drainAll,
   isDraining,
@@ -221,6 +222,7 @@ app.get('/health', c => {
     uptime: getUptime(),
     draining: isDraining() || undefined,
     startedAt: getStartTime(),
+    mayorReadyAt: getMayorReadyAt() ?? undefined,
   };
   return c.json(response);
 });

services/gastown/container/src/process-manager.ts

@@ -78,6 +78,26 @@ export function getStartTime(): string {
   return new Date(startTime).toISOString();
 }
 
+// Timestamp (ISO 8601) of the moment the first mayor agent in this container
+// reached 'running' status. Used by /health so the Town DO can compute
+// container-start-to-mayor-ready latency. Stays null until a mayor is up;
+// survives subsequent mayor exits since the window is measured against the
+// first mayor ready in the container's lifetime.
+let mayorReadyAt: string | null = null;
+
+export function getMayorReadyAt(): string | null {
+  return mayorReadyAt;
+}
+
+function markMayorReadyOnce(): void {
+  if (mayorReadyAt !== null) return;
+  mayorReadyAt = new Date().toISOString();
+  log.info('mayor.ready', {
+    containerUptimeMs: getUptime(),
+    mayorReadyAt,
+  });
+}
+
 async function hydrateDbFromSnapshot(
   agentId: string,
   apiUrl: string,
@@ -1044,6 +1064,9 @@ export async function startAgent(
     // despite being active — causing the drain to wait indefinitely.
     if (agent.status === 'starting') {
       agent.status = 'running';
+      if (request.role === 'mayor') {
+        markMayorReadyOnce();
+      }
     }
 
     // 4. Send the initial prompt

services/gastown/container/src/types.ts

@@ -165,6 +165,10 @@ export type HealthResponse = {
   uptime: number;
   draining?: boolean;
   startedAt?: string;
+  /** ISO 8601 timestamp of the first mayor agent reaching 'running' status
+   *  in this container's lifetime. Used by the worker to measure container
+   *  cold-start → mayor-session-ready latency. */
+  mayorReadyAt?: string;
 };
 
 // ── Kilo serve instance ─────────────────────────────────────────────────