fix(runner): review-loop 소진으로 escalated 도달 시 recordEscalation/notify 누락

자기야 검증: forced REQUEST_CHANGES 로 escalation 까지 갔는데 디스코드 알림이
안 떨어졌음. 로그 분석 결과 recordEscalation() 호출 자체가 0회.

원인: runner.ts 의 main loop 가 recordEscalation 을 ERROR 이벤트의
non-retryable 분기에서만 호출. 그런데 review-loop 소진 + replan 소진으로
FSM 이 자체적으로 escalated 로 전이한 경우는 ERROR 가 아니라 REQUEST_CHANGES
이벤트 경로라서 그 분기를 안 탐. 결과: escalation row 안 만들어지고 notifier
도 안 호출됨.

수정:
- escalationRecorded flag 추가, ERROR non-retryable 분기에서 true 로 세팅
- lastActiveStage 를 매 iteration 마다 트래킹
- main loop 종료 후 result.state === "escalated" && !escalationRecorded 면
  post-loop 에서 recordEscalation() + emit({type:"escalated"}) 호출
- stage 는 lastActiveStage (review-loop 소진의 경우 보통 "review"),
  attempts 는 context.replanCount, reason 은 context.lastError

검증 예정: 다음 forced 테스트에서 디스코드 채널에 escalation SOS 알림 떨어짐.
113 테스트 그대로 통과.

src/orchestrator/runner.ts

@@ -136,6 +136,8 @@ export async function runPipeline(opts: RunOptions): Promise<RunResult> {
   });
 
   let transitions = 1;
+  let escalationRecorded = false;
+  let lastActiveStage: "plan" | "implement" | "review" | "deploy" = "plan";
   const TERMINAL: PipelineState[] = ["done", "escalated", "aborted"];
 
   // Accumulate stage outputs so each stage can see what the previous ones produced.
@@ -159,6 +161,7 @@ export async function runPipeline(opts: RunOptions): Promise<RunResult> {
       log.warn({ state: result.state }, "Non-active state encountered, stopping");
       break;
     }
+    lastActiveStage = stage;
 
     const transport = opts.transports.get(stage);
     if (!transport) {
@@ -247,6 +250,7 @@ export async function runPipeline(opts: RunOptions): Promise<RunResult> {
           },
           opts.notifier,
         );
+        escalationRecorded = true;
         emit({
           type: "escalated",
           pipelineId,
@@ -279,7 +283,46 @@ export async function runPipeline(opts: RunOptions): Promise<RunResult> {
       transitions,
     });
   } else if (result.state === "escalated") {
-    // Escalated emit already fired at the failure site; no extra event.
+    // FSM can reach `escalated` two ways:
+    //   1. ERROR (non-retryable) — recordEscalation was called inline
+    //      and escalationRecorded was set true.
+    //   2. REQUEST_CHANGES exhaustion (review-loop / replan budget) —
+    //      that's a normal handoff event, not an ERROR, so the inline
+    //      branch above never runs. Catch it here.
+    if (!escalationRecorded) {
+      const reason = String(
+        result.context.lastError ?? "Pipeline escalated",
+      );
+      const replanCount = (result.context as { replanCount?: number })
+        .replanCount ?? 0;
+      try {
+        await recordEscalation(
+          {
+            pipelineId,
+            stage: lastActiveStage,
+            reason,
+            attempts: replanCount,
+            contextSnapshot: result.context as unknown as Record<
+              string,
+              unknown
+            >,
+          },
+          opts.notifier,
+        );
+      } catch (err) {
+        log.warn(
+          { err: err instanceof Error ? err.message : String(err) },
+          "post-loop recordEscalation failed",
+        );
+      }
+      emit({
+        type: "escalated",
+        pipelineId,
+        stage: lastActiveStage,
+        reason,
+        attempts: replanCount,
+      });
+    }
   } else {
     emit({
       type: "failed",