fix(gateway): reap cgroup PIDs from a fresh read taken after the live-gateway guard

The single cgroup.procs read now feeds only the empty-check and the
live-gateway guard. The guard can take seconds (gateway.status import,
per-PID cmdline with psutil/ps fallback), so killing from that stale list
could SIGKILL a PID that exited and was reused outside the cgroup, and
would miss orphans spawned in the gap. Re-read cgroup.procs for the kill
loop (one small sysfs read), restoring base's kill-set freshness while
keeping the skipped import on the common empty-cgroup stop.

gateway/cgroup_cleanup.py

@@ -97,7 +97,12 @@ def reap_cgroup(cgroup_path: str | None = None) -> int | None:
         )
         return None
     killed = 0
-    for pid in others:
+    # Kill from a fresh read: the guard above can take seconds (status import,
+    # per-PID cmdline/ps fallback), so `others` may hold exited/reused PIDs and
+    # miss orphans spawned meanwhile.
+    for pid in _read_cgroup_pids(cgroup_path):
+        if pid == me:
+            continue
         try:
             os.kill(pid, signal.SIGKILL)  # windows-footgun: ok — Linux-only (reads /proc, /sys/fs/cgroup; runs from a systemd unit)
             killed += 1

tests/gateway/test_cgroup_cleanup.py

@@ -102,3 +102,12 @@ class TestLiveGatewayGuard:
         monkeypatch.setattr(cgroup_cleanup.os, "kill", lambda pid, sig: killed.append(pid))
         assert cgroup_cleanup.reap_cgroup(cgroup_path) == 1
         assert killed == [777]
+
+        # The kill set comes from a fresh cgroup.procs read taken after the
+        # (slow) guard: a PID that exited meanwhile is not signalled (it may be
+        # reused outside the cgroup) and an orphan spawned meanwhile is reaped.
+        reads = iter([[777, os.getpid()], [888, os.getpid()]])
+        monkeypatch.setattr(cgroup_cleanup, "_read_cgroup_pids", lambda _p: next(reads))
+        killed.clear()
+        assert cgroup_cleanup.reap_cgroup(cgroup_path) == 1
+        assert killed == [888]