feat(server): 工程骨架后端——上传 → 解析 → 清单最小链路

- core：配置（prod 拒绝默认密钥 / dev 短信 / 假模型）、租户事务 + 行级安全策略、
  problem+json 错误码、信封加密与签名下载、对象存储、事务内入队、Valkey stream 事件流、权限点
- identity：手机号 HMAC 存储、注册、登录（挤下线需确认，API3）、服务端会话
- billing：额度流水只追加，预扣 / 结算 / 退回，一笔预扣只有一个结局
- document：分片上传、按文件头识别格式、Gotenberg 转换、文字层与稳定句子编号
- checklist：标注规则识别、模型抽取、三道校验、标注类条款规则兜底、重跑对齐
- ai：提示词注册表（extract.substantive v1 来自 S4 v2）、假模型、OpenAI 兼容供应商、ai_call 流水
- 迁移：data-model 基线、套餐、Procrastinate、会话状态函数
- 测试 30 个：租户隔离、额度、失败退回、对象加密、契约一致性
- Compose（含 migrate）、镜像、CI

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

server/src/bid_doctor/ai/providers/base.py

@@ -0,0 +1,28 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Protocol
+
+
+@dataclass(frozen=True, slots=True)
+class Completion:
+    data: dict[str, Any]
+    raw: str
+    model: str
+    tokens_in: int | None = None
+    tokens_out: int | None = None
+    output_mode: str = "json_mode"
+
+
+class TransientError(Exception):
+    """限流、5xx、网络抖动：可以重试；重试后仍失败按基础设施错误处理（退回额度）。"""
+
+    def __init__(self, error_class: str, message: str) -> None:
+        super().__init__(message)
+        self.error_class = error_class
+
+
+class Provider(Protocol):
+    name: str
+
+    async def complete_json(self, system: str, user: str, capability: str) -> Completion: ...

server/src/bid_doctor/ai/providers/fake.py

@@ -0,0 +1,40 @@
+"""确定性的假模型：开发和测试用，CI 不调用真实模型（架构 §8）。
+
+只认 extract.substantive：把带 ★ 或 <iu> 的句子当作实质性条款，逐字摘一段原文作为引文。
+"""
+
+from __future__ import annotations
+
+import json
+import re
+
+from bid_doctor.ai.providers.base import Completion
+
+LINE = re.compile(r"^\[([^\]]+)\] (.*)$")
+TAG = re.compile(r"</?(iu|u|i)>")
+
+
+class FakeProvider:
+    name = "fake"
+
+    async def complete_json(self, system: str, user: str, capability: str) -> Completion:
+        del system, capability
+        items = []
+        for line in user.split("## 待审分块", 1)[-1].splitlines():
+            m = LINE.match(line.strip())
+            if not m:
+                continue
+            sid, marked = m.groups()
+            plain = TAG.sub("", marked)
+            if "★" in plain:
+                basis = "star"
+            elif "<iu>" in marked:
+                basis = "italic_underline"
+            else:
+                continue
+            if any(k in plain for k in ("标注", "实质性要求", "条款响应")) and len(plain) < 60:
+                continue  # 只是在说明标注规则的句子，不算条款（和提示词的「不要输出」一致）
+            body = plain.lstrip("★*＊ ").strip()
+            items.append({"summary": body[:40], "basis": basis, "evidence": [{"sid": sid, "quote": body[:20]}]})
+        data = {"items": items}
+        return Completion(data=data, raw=json.dumps(data, ensure_ascii=False), model="fake-extractor")

server/src/bid_doctor/ai/providers/openai_compat.py

@@ -0,0 +1,54 @@
+"""OpenAI 兼容协议的国产模型（DeepSeek、通义千问、豆包、GLM、Kimi 等，T1）。
+
+供应商地址和模型由运维在部署配置里设置，用户和租户都不能填（架构 §3.6）。
+TODO(路由表)：按能力配候选列表、按「供应商 × 租户」限流；现在只有一个供应商一个模型。
+"""
+
+from __future__ import annotations
+
+import json
+
+import httpx
+
+from bid_doctor.ai.providers.base import Completion, TransientError
+from bid_doctor.core.config import get_settings
+
+
+class OpenAICompatProvider:
+    name = "openai_compat"
+
+    async def complete_json(self, system: str, user: str, capability: str) -> Completion:
+        del capability
+        s = get_settings()
+        body = {
+            "model": s.ai_model,
+            "temperature": 0,
+            "response_format": {"type": "json_object"},
+            "messages": [{"role": "system", "content": system}, {"role": "user", "content": user}],
+        }
+        try:
+            async with httpx.AsyncClient(base_url=s.ai_base_url, timeout=httpx.Timeout(180, connect=10)) as c:
+                r = await c.post("/chat/completions", json=body, headers={"Authorization": f"Bearer {s.ai_api_key}"})
+        except httpx.TimeoutException as e:
+            raise TransientError("timeout", "模型调用超时") from e
+        except httpx.TransportError as e:
+            raise TransientError("network", "模型调用网络错误") from e
+        if r.status_code == 429:
+            raise TransientError("rate_limited", "模型接口限流")
+        if r.status_code >= 500:
+            raise TransientError("server", f"模型接口 {r.status_code}")
+        r.raise_for_status()
+        payload = r.json()
+        raw = payload["choices"][0]["message"]["content"]
+        usage = payload.get("usage") or {}
+        try:
+            data = json.loads(raw)
+        except json.JSONDecodeError:
+            data = {"items": []}  # 内容质量问题：不算任务失败，相关条目由规则兜底进待确认区（架构 §3.6）
+        return Completion(
+            data=data,
+            raw=raw,
+            model=payload.get("model", s.ai_model),
+            tokens_in=usage.get("prompt_tokens"),
+            tokens_out=usage.get("completion_tokens"),
+        )