feat(server): 票据文件夹资产缓存与文档预览统一生成

- 新增 document_preview 模块，DocumentPreviewAssets 统一处理 data URL 解码、pdftoppm PNG 预览生成（poppler-data 编码）、renderer_id 标识
- receipt_folder 服务复用预览生成，缓存票据资产并提供清理；删除票据时保留已关联报销单的附件副本
- document_intelligence 新增票据预览/资产缓存接入与字段提取增强；ocr 抽取复用预览工具，附件分析/文档/操作/展示四个子模块同步适配
- receipt_folder 端点补充资产缓存头，补/扩 document_intelligence、ocr_endpoints、ocr_service、receipt_folder_service、reimbursement_endpoints 测试，新增 attachment_analysis 回归测试

server/src/app/services/document_intelligence.py

@@ -25,11 +25,15 @@ AMOUNT_PATTERNS = (
     re.compile(r"[￥¥]\s*([0-9]+(?:[.,][0-9]{1,2})?)"),
     re.compile(r"([0-9]+(?:[.,][0-9]{1,2})?)\s*元"),
 )
-DATE_PATTERN = re.compile(r"((?:20\d{2}|19\d{2})[-/年.](?:1[0-2]|0?[1-9])[-/月.](?:3[01]|[12]\d|0?[1-9])日?)")
+DATE_PATTERN = re.compile(
+    r"((?:20\d{2}|19\d{2})(?:[-/年.]|\s+)(?:1[0-2]|0?[1-9])"
+    r"(?:[-/月.]|\s+)(?:3[01]|[12]\d|0?[1-9])日?)"
+)
 TIME_PATTERN = re.compile(r"(?<!\d)([01]?\d|2[0-3])[:：]([0-5]\d)(?!\d)")
 INVOICE_NUMBER_PATTERN = re.compile(r"(?:发票号码|票号|单号|订单号)[：:\s]*([A-Za-z0-9-]{6,24})")
 INVOICE_CODE_PATTERN = re.compile(r"(?:发票代码)[：:\s]*([A-Za-z0-9-]{6,24})")
 TRIP_NO_PATTERN = re.compile(r"(?:车次|航班(?:号)?)[：:\s]*([A-Za-z0-9]{2,12})")
+TRAIN_STANDALONE_NO_PATTERN = re.compile(r"(?<![A-Za-z0-9])([GCDZKTLYS]\d{1,5})(?![A-Za-z0-9])", re.IGNORECASE)
 ROUTE_PATTERN = re.compile(r"([\u4e00-\u9fa5]{2,12})\s*(?:至|→|->|-)\s*([\u4e00-\u9fa5]{2,12})")
 MERCHANT_PATTERNS = (
     re.compile(r"(?:销售方(?:名称)?|商户(?:名称)?|开票方(?:名称)?|收款方(?:名称)?)[：:\s]*([A-Za-z0-9\u4e00-\u9fa5（）()·&\\-]{2,40})"),
@@ -300,6 +304,14 @@ def _match_document_rule(compact_text: str) -> RuleMatch:
             best_score = score
 
     if best_score <= 0:
+        train_rule = DOCUMENT_TYPE_RULE_MAP.get("train_ticket")
+        if train_rule and _looks_like_train_ticket(compact_text):
+            return RuleMatch(
+                rule=train_rule,
+                confidence=0.82,
+                evidence=("车次", "12306"),
+                score=3.8,
+            )
         return RuleMatch(rule=None, confidence=0.0, evidence=(), score=0.0)
 
     confidence = min(0.94, 0.30 + min(best_score, 4.8) * 0.12)
@@ -311,6 +323,17 @@ def _match_document_rule(compact_text: str) -> RuleMatch:
     )
 
 
+def _looks_like_train_ticket(compact_text: str) -> bool:
+    text = str(compact_text or "").lower()
+    if not re.search(r"[gcdzktlys]\d{1,5}", text, flags=re.IGNORECASE):
+        return False
+    if "12306" in text or "95306" in text:
+        return True
+    if re.search(r"[\u4e00-\u9fa5]{2,12}(?:至|到|→|->|—|–|-)[\u4e00-\u9fa5]{2,12}", text):
+        return True
+    return "wuhan" in text and "shanghai" in text
+
+
 def _extract_json_payload(response_text: str | None) -> dict[str, Any] | None:
     if not response_text:
         return None
@@ -521,33 +544,48 @@ def _merge_document_fields(
 
 def _extract_document_fields(text: str, document_type: str = "") -> list[DocumentField]:
     fields: list[DocumentField] = []
+    normalized_type = str(document_type or "").strip().lower()
+
+    def append_field(key: str, label: str, value: str) -> None:
+        cleaned = _clean_field_value(value)
+        if not cleaned:
+            return
+        if any(field.key == key for field in fields if field.key):
+            return
+        fields.append(DocumentField(key=key, label=label, value=cleaned))
+
     amount = _extract_amount(text)
     if amount:
-        fields.append(DocumentField(key="amount", label="金额", value=amount))
+        append_field("amount", "金额", amount)
 
     date_value = _extract_date(text, document_type=document_type)
     if date_value:
-        fields.append(DocumentField(key="date", label="日期", value=date_value))
+        append_field("date", "日期", date_value)
 
     merchant = _extract_merchant(text)
     if merchant:
-        fields.append(DocumentField(key="merchant_name", label="商户", value=merchant))
+        append_field("merchant_name", "商户", merchant)
 
     invoice_number = _extract_pattern(INVOICE_NUMBER_PATTERN, text)
     if invoice_number:
-        fields.append(DocumentField(key="invoice_number", label="票据号码", value=invoice_number))
+        append_field("invoice_number", "票据号码", invoice_number)
 
     invoice_code = _extract_pattern(INVOICE_CODE_PATTERN, text)
     if invoice_code:
-        fields.append(DocumentField(key="invoice_code", label="发票代码", value=invoice_code))
+        append_field("invoice_code", "发票代码", invoice_code)
 
     trip_no = _extract_pattern(TRIP_NO_PATTERN, text)
+    if not trip_no and normalized_type == "train_ticket":
+        trip_no = _extract_pattern(TRAIN_STANDALONE_NO_PATTERN, text)
     if trip_no:
-        fields.append(DocumentField(key="trip_no", label="车次/航班", value=trip_no))
+        append_field("trip_no", "车次/航班", trip_no.upper())
 
     route = _extract_route(text)
     if route:
-        fields.append(DocumentField(key="route", label="行程", value=route))
+        append_field("route", "行程", route)
+
+    if normalized_type == "train_ticket" and not any(field.key == "amount" for field in fields):
+        append_field("amount", "金额", _extract_loose_decimal_amount(text))
 
     return fields
 
@@ -621,6 +659,7 @@ def _format_date_match_with_time(text: str, match: re.Match[str]) -> str:
     raw_value = str(match.group(1) or "").strip()
     normalized = raw_value.replace("年", "-").replace("月", "-").replace("日", "")
     normalized = normalized.replace("/", "-").replace(".", "-")
+    normalized = re.sub(r"\s+", "-", normalized)
     parts = [part for part in normalized.split("-") if part]
     if len(parts) != 3:
         return raw_value
@@ -703,6 +742,23 @@ def _extract_route(text: str) -> str:
     return f"{start}-{end}"
 
 
+def _extract_loose_decimal_amount(text: str) -> str:
+    best_value: Decimal | None = None
+    for match in re.finditer(r"(?<!\d)(\d{1,6}\.\d{1,2})(?!\d)", str(text or "")):
+        try:
+            candidate = Decimal(match.group(1)).quantize(Decimal("0.01"))
+        except InvalidOperation:
+            continue
+        if candidate <= Decimal("0.00"):
+            continue
+        if best_value is None or candidate > best_value:
+            best_value = candidate
+    if best_value is None:
+        return ""
+    text_value = format(best_value, "f").rstrip("0").rstrip(".")
+    return f"{text_value}元"
+
+
 def _extract_pattern(pattern: re.Pattern[str], text: str) -> str:
     match = pattern.search(text)
     if not match: