Treat zombie PIDs as dead in batch update regression

Why: CI containers leave unreaped zombies; os.kill(0) still succeeds
and falsely failed the old-process-stopped assertion.

Evidence: local unittest of the batch update case.

Impact: assertion matches svc.sh/recover zombie semantics.

Refs: .task/feature-svc-pidcheck-leak.md; PR #457
Co-authored-by: Cursor <cursoragent@cursor.com>

tests/test_svc.py

@@ -634,9 +634,16 @@ class SvcDbRecoveryTests(unittest.TestCase):
             new_pid = int((self.runtime(name) / "svc.pid").read_text().strip())
             self.assertNotEqual(new_pid, old_pids[name],
                                 "%s 应已换新进程" % name)
+            # 与 svc.sh / recover 夹具一致：僵尸按死处理（CI 容器 PID 1 不收割）。
+            old_still = False
             try:
                 os.kill(old_pids[name], 0)
-                old_still = True
+                try:
+                    state = Path("/proc/%d/stat" % old_pids[name]).read_text() \
+                        .rsplit(")", 1)[1].split()[0]
+                    old_still = state != "Z"
+                except (OSError, IndexError):
+                    old_still = True
             except (ProcessLookupError, PermissionError):
                 old_still = False
             self.assertFalse(old_still, "%s 旧进程应已停止" % name)