feat(harness): hold recovery while a Grafana alert is firing, add legacy:busy host resolution

キャンペーン第 2 回のグレーゾーン 3 件を受けた変更。SLO 3 系列に現れない症状
（レプリケーション一時停止、メモリ圧迫）でも、受信中のアラートが解決するまで
recovered を出さないよう、AlertRouter が firing 中のアラートを保持し judge が
alerts_ok を条件に加える。legacy:busy は直近 2 分で最も多くリクエストを受けた
MIG インスタンス（Mimir topk）を指し、run+シナリオ単位でキャッシュして revert が
同じインスタンスに当たるようにする（ILB のハッシュで api の接続が 1 台に偏る）。

harness/judge.go

@@ -45,6 +45,35 @@ func (a *App) mimirScalar(ctx context.Context, expr string) (float64, error) {
 	return strconv.ParseFloat(s, 64)
 }
 
+// mimirTopInstance runs an instant query and returns the "instance" label of the first sample.
+func (a *App) mimirTopInstance(ctx context.Context, expr string) (string, error) {
+	u := a.cfg.MimirURL + "/api/v1/query?query=" + url.QueryEscape(expr)
+	req, _ := http.NewRequestWithContext(ctx, "GET", u, nil)
+	resp, err := (&http.Client{Timeout: 15 * time.Second}).Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	b, _ := io.ReadAll(resp.Body)
+	if resp.StatusCode != 200 {
+		return "", fmt.Errorf("mimir %s: %s", resp.Status, truncate(string(b), 300))
+	}
+	var r struct {
+		Data struct {
+			Result []struct {
+				Metric map[string]string `json:"metric"`
+			} `json:"result"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(b, &r); err != nil {
+		return "", err
+	}
+	if len(r.Data.Result) == 0 || r.Data.Result[0].Metric["instance"] == "" {
+		return "", fmt.Errorf("no instance sample for %s", expr)
+	}
+	return r.Data.Result[0].Metric["instance"], nil
+}
+
 // SLOState is one evaluation of experiment design 4.2 (probe 12 consecutive successes = min over 3m; p95 and 5xx ratio over 5m).
 // Values are nil when the series is absent (JSON has no NaN).
 // 2026-10-08 (campaign 2, C-G1): a 300 ms delay that lasted 3 minutes was declared recovered while still active because the
@@ -59,6 +88,8 @@ type SLOState struct {
 	LatencyOK    bool      `json:"latency_ok"`
 	LatencyNowOK bool      `json:"latency_1m_ok"`
 	ErrorsOK     bool      `json:"errors_ok"`
+	AlertsFiring []string  `json:"alerts_firing,omitempty"` // Grafana alerts received firing and not yet resolved (2026-10-08: grey-zone symptoms live outside the three SLO series)
+	AlertsOK     bool      `json:"alerts_ok"`
 	Recovered    bool      `json:"recovered"`
 	MissingData  []string  `json:"missing_data,omitempty"`
 	QueryErrors  []string  `json:"query_errors,omitempty"`
@@ -93,6 +124,10 @@ func (a *App) evaluateSLO(ctx context.Context) SLOState {
 	st.LatencyNowOK = st.P95Now == nil || *st.P95Now < 0.5
 	st.ErrorRatio = get("errors", qErrRatio)
 	st.ErrorsOK = st.ErrorRatio == nil || *st.ErrorRatio < 0.01
-	st.Recovered = st.ProbeOK && st.LatencyOK && st.LatencyNowOK && st.ErrorsOK && len(st.QueryErrors) == 0
+	if a.hooks != nil {
+		st.AlertsFiring = a.hooks.Firing()
+	}
+	st.AlertsOK = len(st.AlertsFiring) == 0
+	st.Recovered = st.ProbeOK && st.LatencyOK && st.LatencyNowOK && st.ErrorsOK && st.AlertsOK && len(st.QueryErrors) == 0
 	return st
 }