fix(gastown): check dispatchAgent return value instead of using .catch()

dispatchAgent resolves false (never rejects) on container start failure,
so the .catch() handler was dead code. Changed to check the boolean
return value so the idle rollback actually executes on failure.

Author: John Fawcett

cloudflare-gastown/src/dos/town/actions.ts

@@ -558,17 +558,19 @@ export function applyAction(ctx: ApplyActionContext, action: Action): (() => Pro
 
       const capturedAgentId = agentId;
       return async () => {
-        // Best-effort dispatch. On failure, roll the agent back to 'idle'
-        // immediately so the reconciler can retry dispatch on the next tick
-        // without waiting for heartbeat timeout (90s). The bead stays
-        // 'in_progress' so it remains eligible for re-dispatch. (§5.4)
-        await ctx.dispatchAgent(capturedAgentId, beadId, rigId).catch(err => {
+        // Best-effort dispatch. dispatchAgent resolves false (never rejects)
+        // on failure, so we check the return value. On failure, roll the
+        // agent back to 'idle' immediately so the reconciler can retry
+        // dispatch on the next tick without waiting for heartbeat timeout
+        // (90s). The bead stays 'in_progress' so it remains eligible for
+        // re-dispatch. (§5.4)
+        const started = await ctx.dispatchAgent(capturedAgentId, beadId, rigId);
+        if (!started) {
           console.warn(
-            `${LOG} dispatch_agent: container start failed for agent=${capturedAgentId} bead=${beadId}, rolling back to idle`,
-            err
+            `${LOG} dispatch_agent: container start failed for agent=${capturedAgentId} bead=${beadId}, rolling back to idle`
           );
           agentOps.updateAgentStatus(sql, capturedAgentId, 'idle');
-        });
+        }
       };
     }