fix: serialize concurrent PG schema init with advisory lock

Startup race: concurrent init_db (cloud-server + worker) ran
CREATE TABLE / CREATE INDEX IF NOT EXISTS against a DB where the
object did not exist yet; the check-then-create is not atomic, so
both sessions created it and one died with UniqueViolation on
pg_type_typname_nsp_index (tables) or pg_class_relname_nsp_index
(indexes). Hit on the 37ba687 and 7983567 deploys (7983567 added
idx_tasks_parent); systemd restart healed it every time.

Fix (advisory lock, not catch-duplicate_object): init_postgres_schema
takes pg_advisory_xact_lock (key 20261108, released at its commit),
and server.init_db takes the session-level lock for the whole PG
branch so the _ensure_* objects outside POSTGRES_SCHEMA
(task_continue_messages, user_gateway_keys, task_attempts,
task_audit_loop_state) are covered too. Same idiom as the existing
RATE_LIMIT_SWEEP_ADVISORY_KEY sweep lock; key is distinct.

Tests: two new real-PG regression tests (skipped without
CLOUD_TEST_PG_DSN), 4 processes racing on an empty schema, for
init_postgres_schema and for full init_db; both fail on the
unfixed code and pass with the fix.

cloud/tests/test_postgres_backend.py

@@ -17,7 +17,9 @@
 import os
 import socket
 import sys
+import traceback
 import unittest
+from multiprocessing import Barrier, Process, Queue
 from pathlib import Path
 from unittest import mock
 from urllib.parse import unquote, urlparse
@@ -386,3 +388,81 @@ class RealPostgresSchemaTestCase(unittest.TestCase):
             except Exception:
                 pass
             conn.close()
+
+
+def _concurrent_schema_worker(barrier, queue, wid, params):
+    """并发初始化子进程：与其它进程同时跑 init_postgres_schema。"""
+    try:
+        host, port, dbname, user, password = params
+        barrier.wait(timeout=60)
+        db = PostgresDatabase(host=host, port=port, dbname=dbname,
+                              user=user, password=password)
+        conn = db.connect()
+        try:
+            init_postgres_schema(conn)
+        finally:
+            conn.close()
+        queue.put((wid, "ok", ""))
+    except Exception:
+        queue.put((wid, "FAIL", traceback.format_exc()))
+
+
+class RealPostgresConcurrentInitTestCase(unittest.TestCase):
+    """真实 PG 并发初始化：无 DSN 或不可达时 skip。
+
+    回归：启动竞态（37ba687 与 7983567 两次部署均撞上）——
+    cloud-server 与 worker 并发 init 时，CREATE TABLE / CREATE
+    INDEX 的 IF NOT EXISTS 检查与建对象不是原子的，并发建同名
+    对象会撞 pg_type_typname_nsp_index（建表）或
+    pg_class_relname_nsp_index（建索引）的 UniqueViolation。
+    修法是 schema 初始化持 advisory 锁串行化；本用例在空库上
+    4 进程同时初始化，必须全部成功、表齐全、无一例异常外泄。
+    """
+
+    def test_concurrent_init_postgres_schema_no_unique_violation(self):
+        params = _real_pg_params()
+        if params is None:
+            self.skipTest("未设置可达的 CLOUD_TEST_PG_DSN，跳过真实 PG 测试")
+        if not _has_real_driver():
+            self.skipTest("未装 psycopg 驱动，跳过真实 PG 测试")
+        host, port, dbname, user, password = params
+        db = PostgresDatabase(host=host, port=port, dbname=dbname,
+                              user=user, password=password)
+        conn = db.connect()
+        try:
+            conn.execute("DROP SCHEMA public CASCADE")
+            conn.execute("CREATE SCHEMA public")
+            conn.commit()
+        finally:
+            conn.close()
+        nworkers = 4
+        barrier = Barrier(nworkers)
+        queue = Queue()
+        procs = [Process(target=_concurrent_schema_worker,
+                         args=(barrier, queue, i, params))
+                 for i in range(nworkers)]
+        try:
+            for proc in procs:
+                proc.start()
+            for proc in procs:
+                proc.join(timeout=120)
+        finally:
+            for proc in procs:
+                if proc.is_alive():
+                    proc.terminate()
+        results = [queue.get() for _ in range(nworkers)]
+        failures = [r for r in results if r[1] == "FAIL"]
+        self.assertEqual([], failures,
+                         "并发初始化有进程失败：%s" % (failures,))
+        conn = db.connect()
+        try:
+            rows = db.read_all(
+                conn,
+                "SELECT table_name FROM information_schema.tables"
+                " WHERE table_schema = 'public'",
+                ())
+            names = {r["table_name"] for r in rows}
+            for table in SchemaStatementTestCase.EXPECTED_TABLES:
+                self.assertIn(table, names)
+        finally:
+            conn.close()