fix(deploy): 启动脚本等端口真正释放，并确认服务可用后才报成功

上一次部署里旧 worker 没随 master 退出，一直占着 8123。新 master 启动、
打印 started、然后什么都不服务——日志每一行看起来都正常，站点却是挂的。

- stop.sh 等进程真正退出，超时后 -9
- start.sh 停完等端口释放，起来后 curl 健康检查通过才算成功，否则返回非零

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

deploy/stop.sh

@@ -0,0 +1,19 @@
+#!/bin/sh
+APP="$(cd "$(dirname "$0")" && pwd)"
+GUNICORN="$APP/backend/.venv/bin/gunicorn"
+
+if [ -f "$APP/app.pid" ]; then
+  kill "$(cat "$APP/app.pid")" 2>/dev/null && echo "stopped $(cat "$APP/app.pid")"
+  rm -f "$APP/app.pid"
+fi
+
+# The master's workers do not always go with it, and a surviving worker keeps
+# :8123 bound — which looks exactly like a healthy start that serves nothing.
+i=0
+while [ $i -lt 15 ]; do
+  pgrep -f "$GUNICORN" >/dev/null 2>&1 || exit 0
+  sleep 1
+  i=$((i + 1))
+done
+pkill -9 -f "$GUNICORN" 2>/dev/null
+echo "force-stopped leftover workers"