perf(backend): 内存优化三项——全市场研究不再占满 8G

1) 数据装配流式+列裁剪：Repository 新增 stream_range_many_columns（只 SELECT
   所需列、SQL 侧转 REAL、yield_per 分批），引擎按 required_columns 取数
   （LocalEngine 仅 close+因子字段），消除 ORM/Decimal 全量物化；
2) 研究 Job 独立子进程执行（job.mode=subprocess）：python -m app.cli.run_job
   在子进程内设 RLIMIT_AS 上限，OOM 归档 failed 而非拖垮 API worker；
   子进程异常退出由父进程补记 failed；并发上限 2；
3) 服务启动清理：残留 queued/running Job 标记 failed（防永久 running）。

实测同款全市场回测：uvicorn worker RSS 稳定 ~220MB，任务峰值内存由 4.1GB+
降至 ~470MB，24s 完成并归档（此前 43s 未完成即 OOM）。
新增/更新测试 96 passed，ruff 干净。

backend/tests/test_jobs_experiments.py

@@ -2,9 +2,11 @@
 
 from __future__ import annotations
 
+import sys
 from datetime import date, datetime
 
 import pytest
+from app.application.services import job_executor as je
 from app.application.services.job_executor import execute_job
 from app.domain.entities.research import (
     JobRecord,
@@ -147,6 +149,64 @@ class TestJobExecutor:
             assert done.error
 
 
+class TestJobResilience:
+    """内存隔离专项：启动残留清理 + 子进程异常退出兜底。"""
+
+    def test_mark_stale_jobs_failed(self, sf) -> None:
+        repo_f = lambda s: SqlAlchemyJobRepository(s)  # noqa: E731
+        rows = [
+            ("JOB-STALE-1", JobStatus.RUNNING),
+            ("JOB-STALE-2", JobStatus.QUEUED),
+            ("JOB-OK-3", JobStatus.SUCCESS),
+        ]
+        for jid, status in rows:
+            job = JobRecord(
+                id=jid,
+                kind="backtest",
+                spec_json=_spec().model_dump_json(),
+                status=status,
+                created_at=datetime.now(),
+            )
+            with sf() as session:
+                repo_f(session).create(job)
+                session.commit()
+
+        n = je.mark_stale_jobs_failed(session_factory=sf, job_repo_factory=repo_f)
+        assert n == 2
+        with sf() as session:
+            stale = repo_f(session).get("JOB-STALE-1")
+            queued = repo_f(session).get("JOB-STALE-2")
+            ok = repo_f(session).get("JOB-OK-3")
+        assert stale is not None and stale.status == JobStatus.FAILED
+        assert stale.finished_at is not None and "重启" in (stale.error or "")
+        assert queued is not None and queued.status == JobStatus.FAILED
+        assert ok is not None and ok.status == JobStatus.SUCCESS
+
+    def test_subprocess_abnormal_exit_marks_failed(self, sf, monkeypatch) -> None:
+        repo_f = lambda s: SqlAlchemyJobRepository(s)  # noqa: E731
+        job = JobRecord(
+            id="JOB-SUB-1",
+            kind="backtest",
+            spec_json=_spec().model_dump_json(),
+            status=JobStatus.QUEUED,
+            created_at=datetime.now(),
+        )
+        with sf() as session:
+            repo_f(session).create(job)
+            session.commit()
+        monkeypatch.setattr(je, "_job_mode", lambda: "subprocess")
+        monkeypatch.setattr(
+            je, "_job_worker_cmd", lambda job_id: [sys.executable, "-c", "import sys; sys.exit(7)"]
+        )
+        je.run_job_background("JOB-SUB-1", session_factory=sf, job_repo_factory=repo_f)
+
+        with sf() as session:
+            done = repo_f(session).get("JOB-SUB-1")
+        assert done is not None
+        assert done.status == JobStatus.FAILED
+        assert done.error is not None and "code=7" in done.error
+
+
 @pytest.fixture()
 def seeded_api_db(tmp_path, monkeypatch) -> None:
     """把全局 SessionLocal 指向 tmp 种子库（5 只股票 × 300 交易日），