feat: 读标补齐五类清单与项目基本信息，修复国动样本暴露的 4 个问题

- 文字层 v2：页眉页脚带单独成句；写到页尾的句子和下一页首句接续（出处两页各一条）
- 结构识别（最小版）：十进制编号、冒号引出的列表找上级条款；上级条款写明无效的，条目照样认定
- 抽取：资格、评分、日期与递交、技术参数、项目基本信息各一份提示词；实质性条款出 v2
  （认上级条款、补「做无效响应处理」等写法）；各类取值核对；按首句合并；
  拆开被合成一条的列举条款；描述标注方式的句子和表头碎片不当条款；* 号标注
- 项目：解析出的名称、编号、采购人、预算、截止时间、分包写回项目，用户改过的不覆盖
- 前端：清单项显示分值、参数要求、日期、证明材料
- 测试 40 个，新增两页虚构样例

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

server/tests/test_parse_flow.py

@@ -159,3 +159,44 @@ async def test_objects_are_encrypted_at_rest(user: User) -> None:
     assert files and not any(p.read_bytes().startswith(b"%PDF") for p in files)
     plain = (FIXTURES / "tender-mini.pdf").read_bytes()[:64]
     assert not any(plain in p.read_bytes() for p in files)
+
+
+async def test_all_categories_and_project_info(user: User) -> None:
+    """虚构的两页样例：项目基本信息、截止时间、资格、评分、技术参数、上级标题下的无效情形、* 号兜底、页脚不进清单。"""
+    project_id, doc_id = await project_with_tender(user, "tender-full.pdf")
+    job = await run_parse(user, doc_id)
+    assert job["status"] == "succeeded", job
+
+    project = (await user.get(f"/projects/{project_id}")).json()
+    assert project["name"] == "某市智慧政务服务平台建设项目"  # 建项目时没填名称：按招标文件自动填
+    assert project["code"] == "TEST-2026-001"
+    assert project["purchaser"] == "某市行政审批局"
+    assert project["legal_regime"] == "gov_procurement"
+    assert project["deadline_at"].startswith("2026-12-01T01:30")  # 北京时间 9:30
+    fields = {f["key"]: f for f in project["fields"]}
+    assert fields["budget_total"]["value"] == "450000.00"
+    assert fields["budget_total"]["citation"]["page_no"] == 1
+
+    items = (await user.get(f"/projects/{project_id}/checklist/items")).json()["items"]
+    by_text = {i["requirement_text"]: i for i in items}
+
+    def item(prefix: str) -> dict[str, object]:
+        return next(v for k, v in by_text.items() if k and k.startswith(prefix))
+
+    for clause in ("25.3.1", "25.3.2"):  # 本身没写「无效」，靠上级标题认定
+        it = item(clause)
+        assert it["zone"] == "formal" and "invalidity" in it["categories"], it
+    assert item("资格要求")["categories"] == ["qualification"]
+    assert item("技术方案")["spec"]["scoring"]["points"] == "20"
+    assert item("系统支持")["spec"]["tech_param"] == {"param": "系统支持", "op": ">=", "value": "200", "unit": "个"}
+    deadline = item("响应文件递交截止时间")
+    assert deadline["spec"]["schedule"]["date_kind"] == "bid_deadline"
+    star = item("*供应商须提供原厂授权书")  # 假模型不认 * 号：规则兜底进待确认
+    assert (star["zone"], star["pending_reason"], star["importance"]) == ("pending", "marker_uncovered", "substantive")
+    assert not any(k and ("—1—" in k or "—2—" in k) for k in by_text)  # 页码不混进条款
+    assert not any(k and "斜体且有下划线" in k for k in by_text)  # 规则说明本身不是条款
+
+    checklist = (await user.get(f"/projects/{project_id}/checklist")).json()
+    assert {r["marker"] for r in checklist["marking_rules"]} == {"italic_underline", "asterisk"}
+    sentences = (await user.get(f"/documents/{doc_id}/pages/1/sentences")).json()["items"]
+    assert [s["location"] for s in sentences if s["text"] == "—1—"] == ["header_footer"]