Retry the e2e login while the server is coming up

Why: one isolated run reached a listening server and still failed the single login attempt, while the same data accepted the password a moment later.
Evidence: replaying the login against that data directory returned HTTP 200.
Impact: the seed retries login for a few seconds before it stops. The password and the set-password step are unchanged.
Refs: scripts/e2e.sh
Co-authored-by: Cursor <cursoragent@cursor.com>

scripts/e2e.sh

@@ -124,9 +124,16 @@ stop_serve() {
   [ -n "${API_PID:-}" ] && { kill "$API_PID" 2>/dev/null || true; wait "$API_PID" 2>/dev/null || true; API_PID=""; }
 }
 login() {
-  curl -sf -c "$JAR" -X POST "$API_BASE/api/auth/login" \
-    -H 'Content-Type: application/json' \
-    -d "{\"owner\":\"$OWNER\",\"password\":\"$PASSWORD\"}" >/dev/null || die "login failed"
+  local i
+  for i in $(seq 1 20); do
+    if curl -sf -c "$JAR" -X POST "$API_BASE/api/auth/login" \
+      -H 'Content-Type: application/json' \
+      -d "{\"owner\":\"$OWNER\",\"password\":\"$PASSWORD\"}" >/dev/null; then
+      return 0
+    fi
+    sleep 0.25
+  done
+  die "login failed"
 }
 # CLI 建号留下 must_change_password。会话在改密前不能调用业务命令。
 # 强制设密会作废当前会话，所以设密后用新密码再登录。