fix(deploy): 延长数据库副本重同步门限

CNPG 主库在 5 秒内晋升，但跨地域 pg_rewind 需要重传约 1.27GB，原 6 分钟等待不足。将同步副本恢复等待扩展到 30 分钟，并增加退出时清理写入探针的保护。\n\n已通过 bash -n、ShellCheck 和 git diff --check。

scripts/cluster/failure-drill.sh

@@ -51,10 +51,16 @@ case $drill in
     cluster_ssh "$CLUSTER_NINGBO_HOST" bash -s -- "$new_primary" <<'REMOTE'
 set -euo pipefail
 primary=$1
+cleanup_probe() {
+  k3s kubectl exec -n easyai "$primary" -c postgres -- \
+    psql -X -U postgres -d easyai_ai_gateway -c \
+    "DELETE FROM system_settings WHERE setting_key='ha_acceptance_probe';" >/dev/null 2>&1 || true
+}
+trap cleanup_probe EXIT HUP INT TERM
 k3s kubectl exec -n easyai "$primary" -c postgres -- \
   psql -X -v ON_ERROR_STOP=1 -U postgres -d easyai_ai_gateway -c \
   "INSERT INTO system_settings(setting_key,value) VALUES('ha_acceptance_probe',jsonb_build_object('at',now())) ON CONFLICT(setting_key) DO UPDATE SET value=excluded.value,updated_at=now();" >/dev/null
-for _ in $(seq 1 180); do
+for _ in $(seq 1 900); do
   if [[ $(k3s kubectl get cluster easyai-postgres -n easyai -o jsonpath='{.status.readyInstances}') == 2 ]]; then
     replication=$(k3s kubectl exec -n easyai "$primary" -c postgres -- \
       psql -X -U postgres -d easyai_ai_gateway -At -c \
@@ -67,6 +73,7 @@ done
 k3s kubectl exec -n easyai "$primary" -c postgres -- \
   psql -X -v ON_ERROR_STOP=1 -U postgres -d easyai_ai_gateway -c \
   "DELETE FROM system_settings WHERE setting_key='ha_acceptance_probe';" >/dev/null
+trap - EXIT HUP INT TERM
 REMOTE
     echo "database_failure_drill=PASS promotion_seconds=$elapsed old_primary=$old_primary new_primary=$new_primary"
     ;;