feat: 查标第 1 步——上传投标分册 → 检查 → 逐条处理 → 复查沿用

后端
- check 模块：检查预览、发起、进度、取消、记录；六个阶段（准备投标文件、响应匹配、规则、证据校验、
  状态沿用、汇总）；问题列表、处理、撤销、批量处理（只限提示级）。
- 每条清单项的检查结论 check_item_result（C9，迁移 0005）：「已响应」也带投标原文出处；
  没有校验通过的证据不能判已响应（C3）。级别从清单项算，不让模型定；说明不写整改建议。
- 响应匹配先只有第三层：内存 BM25 检索候选句 + check.match v1 提示词，检索不到不调模型，
  漏答记 missing_answer，模型临时错误只影响那一批。
- 计费：免费版不记流水；专业版 24 小时内复查免扣（记一行 0）；expected_charge 不一致时拒绝不扣点；
  失败、取消退回；投标文件读不出文字按输入问题失败、不扣点。
- 投标分册：类型（volume_kind）和版本；电子投标平台加密格式在上传时拦下。
- 解析流水线前几个阶段拆到 document/pipeline/normalize.py，解析和检查共用。
- ops 模块：反馈接口；致命问题确认无误的说明自动进误报反馈（U4）。
- SK1：Valkey 客户端短超时、SSE 连接池上限。SK2：CI 加 Gotenberg、直接跑 init-roles.sh。

前端
- 上传投标文件并检查页、查标页（进度、覆盖范围、按级别筛选、逐条处理和行内说明、撤销、原文抽屉、
  已处理、本次已解决）；项目头抽成公共组件，加「查标」标签页；工作台「下一步」覆盖查标阶段。

契约、文档：openapi.yaml 加 item-results、volume_kind、coverage 字段；data-model、api-design §10、
架构 §8 / §13（SK14）、AGENTS.md 同步。

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

server/tests/test_check_units.py

@@ -0,0 +1,79 @@
+"""查标的纯函数：检索、投标侧证据校验、问题级别（m3-plan §2）。"""
+
+from __future__ import annotations
+
+from uuid import uuid4
+
+from bid_doctor.check import findings, match
+from bid_doctor.check.retrieval import Index, tokens
+from bid_doctor.document.pipeline.text_layer import Sentence
+
+
+def sent(seq: int, text: str, page: int = 1) -> Sentence:
+    return Sentence(key=f"doc:p{page}:s{seq}", page_no=page, seq=seq, text=text, quads=[[0, 0, 1, 1]])
+
+
+def test_tokens_keep_number_unit_and_model_names() -> None:
+    t = tokens("服务器内存不低于32GB，支持10GE和硬件BFD")
+    assert "32gb" in t and "10ge" in t and "bfd" in t and "内存" in t
+
+
+def test_bm25_ranks_the_responding_sentence_first() -> None:
+    index = Index(
+        [
+            sent(1, "我公司提供7×24小时售后服务。"),
+            sent(2, "服务器内存配置为64GB，并提供产品彩页。"),
+            sent(3, "交换机支持硬件BFD，最小发送间隔3.3毫秒。"),
+        ]
+    )
+    (top, *_) = index.search("交换机支持硬件BFD 最小报文发送间隔不大于50毫秒", 3)
+    assert top.sentence.seq == 3
+    assert index.search("完全无关的词语组合", 3) == []
+
+
+def bid() -> tuple[match.BidDoc, dict[str, tuple[match.BidDoc, Sentence]]]:
+    s = sent(9, "服务器内存不低于32GB的要求：我方服务器内存配置为64GB。")
+    d = match.BidDoc(uuid4(), 1, "技术分册", [s])
+    return d, {"d1p1s9": (d, s)}
+
+
+def test_responded_needs_verified_evidence() -> None:
+    """模型说已响应，但证据没通过：不能判已响应，记「未找到响应」并带原因码（C3）。"""
+    _, allowed = bid()
+    ok = match.judge(
+        {"verdict": "responded", "evidence": [{"sid": "d1p1s9", "quote": "我方服务器内存配置为64GB。"}]}, allowed, []
+    )
+    assert ok.verdict == "responded" and ok.evidence_status == "verified" and len(ok.evidence) == 1
+
+    no_ev = match.judge({"verdict": "responded", "evidence": []}, allowed, [])
+    assert (no_ev.verdict, no_ev.evidence_status, no_ev.reason_code) == ("not_found", "unverified", "evidence_missing")
+
+    rewritten = match.judge(
+        {"verdict": "responded", "evidence": [{"sid": "d1p1s9", "quote": "内存为64GB，满足要求。"}]}, allowed, []
+    )
+    assert rewritten.reason_code == "quote_not_original" and rewritten.detail["claimed_pages"] == [1]
+
+    other = match.judge({"verdict": "responded", "evidence": [{"sid": "d1p2s1", "quote": "x" * 10}]}, allowed, [])
+    assert other.reason_code == "sid_unknown"
+
+    missing = match.judge(None, allowed, ["技术分册"])
+    assert (missing.verdict, missing.reason_code) == ("undetermined", "missing_answer")
+
+
+def test_severity_comes_from_the_item_not_the_model() -> None:
+    assert findings.severity({"importance": "substantive", "categories": ["tech_param", "invalidity"]}) == "fatal"
+    assert findings.severity({"importance": "normal", "categories": ["qualification"]}) == "fatal"
+    assert findings.severity({"importance": "important", "categories": ["tech_param"]}) == "important"
+    assert findings.severity({"importance": "normal", "categories": ["scoring"]}) == "important"
+    assert findings.severity({"importance": "normal", "categories": ["tech_param"]}) == "hint"
+
+
+def test_no_value_is_a_hint_and_unverified_is_explained() -> None:
+    item = {"title": "服务器内存", "importance": "substantive", "categories": ["tech_param"]}
+    nv = findings.spec(match.Result("no_value", "model", "verified"), item)
+    assert nv is not None and nv.severity == "hint" and nv.title.startswith("未给出具体值")
+    uv = findings.spec(
+        match.Result("not_found", "model", "unverified", "quote_not_original", detail={"claimed_pages": [3]}), item
+    )
+    assert uv is not None and uv.severity == "fatal" and "第 3 页" in uv.explanation and "没能核对" in uv.explanation
+    assert findings.spec(match.Result("responded", "model", "verified"), item) is None