fix(chv): cell 不再把 /run/chv 的组改回 root

chv-cell 每次跑 cell 都 _ensure_bind_source(BROKER_DIR, uid=0,
gid=0)，把 /run/chv 的组改回 root，抹掉 tmpfiles 的 chv-browser
组；之后 chv-browser 重启时 os.remove(browser.sock) 报
Permission denied、无限 crash-loop（构建 #8 客机 62 次重启才
被发现，03:42 驱动两次跑 cell 正好在补丁后、重启前）。改：组
改取 chv-browser 的 gid（不存在回落 0）；serve() 删 socket
失败时报错说清修法，不再只剩裸 traceback。

cloud/tests/test_chv_opt_stage3.py

@@ -29,6 +29,7 @@ def _load_script(name):
 
 
 BROWSER = _load_script("chv-browser.py")
+CELL = _load_script("chv-cell.py")
 EXECD = _load_script("chv-execd.py")
 
 
@@ -244,6 +245,76 @@ class SnapshotApiTest(unittest.TestCase):
             self.assertEqual(snap["a11y"]["role"], "WebArea")
 
 
+class CellBrokerDirTest(unittest.TestCase):
+    def test_broker_dir_gid_prefers_chv_browser_group(self):
+        fake = mock.Mock()
+        fake.gr_gid = 991
+        with mock.patch.object(CELL.grp, "getgrnam",
+                               return_value=fake) as g:
+            self.assertEqual(CELL._broker_dir_gid(), 991)
+            g.assert_called_once_with("chv-browser")
+
+    def test_broker_dir_gid_falls_back_to_root_when_group_missing(self):
+        with mock.patch.object(CELL.grp, "getgrnam",
+                               side_effect=KeyError("chv-browser")):
+            self.assertEqual(CELL._broker_dir_gid(), 0)
+
+    def test_ensure_bind_source_chowns_to_given_ids(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            CELL._ensure_bind_source(tmp, uid=os.getuid(),
+                                     gid=os.getgid())
+            st = os.stat(tmp)
+            self.assertEqual((st.st_uid, st.st_gid),
+                             (os.getuid(), os.getgid()))
+
+    def test_run_cell_keeps_browser_group_on_broker_dir(self):
+        # 接线钉死：run_cell 传给 broker 目录的 gid 必须是
+        # _broker_dir_gid()，绝不能再写死 gid=0（2026-11-08 live
+        # 根因：gid=0 每次跑 cell 都把 /run/chv 改回 root:root，
+        # chv-browser 下次重启删不掉旧 socket、无限 crash-loop）。
+        src = (ROOT / "chv-guest" / "chv-cell.py").read_text()
+        self.assertIn("_ensure_bind_source(BROKER_DIR, uid=0, "
+                      "gid=_broker_dir_gid())", src)
+        self.assertNotIn("_ensure_bind_source(BROKER_DIR, uid=0, "
+                         "gid=0)", src)
+
+
+class StaleSocketTest(unittest.TestCase):
+    def test_remove_stale_socket_removes_existing_file(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            sock = os.path.join(tmp, "browser.sock")
+            open(sock, "w").write("stale")
+            BROWSER._remove_stale_socket(sock)
+            self.assertFalse(os.path.exists(sock))
+
+    def test_remove_stale_socket_missing_is_noop(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            BROWSER._remove_stale_socket(os.path.join(tmp, "nope.sock"))
+
+    def test_remove_stale_socket_permission_error_names_fix(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            sub = os.path.join(tmp, "run-chv")
+            os.makedirs(sub)
+            sock = os.path.join(sub, "browser.sock")
+            open(sock, "w").write("stale")
+            os.chmod(sub, 0o555)
+            try:
+                # 以 root 跑单测时目录权限挡不住，用 mock 钉死分支。
+                if os.geteuid() == 0:
+                    with mock.patch.object(
+                            BROWSER.os, "remove",
+                            side_effect=PermissionError(13, "deny")):
+                        with self.assertRaises(PermissionError) as ctx:
+                            BROWSER._remove_stale_socket(sock)
+                else:
+                    with self.assertRaises(PermissionError) as ctx:
+                        BROWSER._remove_stale_socket(sock)
+                self.assertIn("0775 root:chv-browser", str(ctx.exception))
+                self.assertIn("chv.conf", str(ctx.exception))
+            finally:
+                os.chmod(sub, 0o755)
+
+
 class ScreenshotTest(unittest.TestCase):
     def test_jpeg_q70_width_1280(self):
         with tempfile.TemporaryDirectory() as tmp: