feat: OCR 服务 + 会议材料模块

ry-ocr/ (新)
  本地发票识别微服务 (PaddleOCR 3.x + FastAPI, 8801)
  - QR 优先: 扫到二维码即取开票时间/发票号/金额; 没扫到/格式不合法直接判非发票, 不跑 OCR
  - 配置 QR_FULL_OCR 控制快路径(false, 0.2s)还是全字段(true, 4.5s)
  - /recognize/invoice (multipart) + /recognize/invoice/by-path (本地路径, 白名单) + /recognize/text
  - is_invoice / from_qr / qr_raw / qr_error / error_code 字段
  - 12 字段发票抽取 (regex + 启发式, 左右主体识别)
  - 超时保护 (15s 单页 / 60s 总流程) + PaddleOCR 单例 + ThreadPoolExecutor

ry-api/ruoyi-business/
  - pom.xml: 加 hutool-http/json/core 5.8.27, lombok 1.18.30 (OcrClient @Slf4j 所需)
  - ocr/: OcrClient + InvoiceResult/Fields/Line + ZipExtractor + InvoiceOcrScheduler
  - oss/: OssUploader + OssConfMeta (OCR 识别后重传 OSS)
  - config/: OcrConfig + OcrExecutorConfig (后台线程池)
  - service/impl/InvoiceOcrService: 后台提交 OCR, ZIP 路径解压识别, 替换场景先清旧
  - 会议材料 CRUD 全套 (BizMeetingAuditLog/Executor/Invoice/Material/Supervisor):
    controller + service + mapper + domain + xml

ry-vue3/
  - MeetingDetail.vue (新建): 会议详情页 (含评分维度章节, 改只读)
  - Meetings.vue / OssFileUploader.vue / router / Login.vue: 适配新字段

ry-api/ruoyi-admin/
  - RuoYiApplication.java + application.yml: 启用 @Async 异步支持

_self/
  - manager_meetings.md / manager_meeting_detail.md: 文档

ry-ocr/tests/__init__.py

@@ -0,0 +1 @@
+"""tests"""

ry-ocr/tests/bench.py

@@ -0,0 +1,44 @@
+"""基准测试：模型预热后，统计 PNG / PDF 单次识别耗时"""
+import sys
+import time
+from pathlib import Path
+
+try:
+    sys.stdout.reconfigure(encoding="utf-8")
+except Exception:
+    pass
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+from app.core import warmup
+from app.services.recognize_service import recognize_file
+
+
+def bench(label: str, file_path: Path, runs: int = 5):
+    content = file_path.read_bytes()
+    print(f"\n=== {label}: {file_path.name} ({len(content)} bytes) ===")
+
+    # 预热（不计耗时） - 现在 warmup 会真跑一次 dummy 图预测
+    print("预热中（不计入）...", flush=True)
+    warmup()
+    print("预热完成", flush=True)
+
+    times = []
+    for i in range(runs):
+        t0 = time.time()
+        result = recognize_file(file_path.name, content)
+        elapsed = (time.time() - t0) * 1000
+        times.append(elapsed)
+        print(f"  第 {i+1} 次: {elapsed:7.1f} ms  success={result.success}")
+
+    times.sort()
+    print(f"\n  min:    {times[0]:7.1f} ms")
+    print(f"  median: {times[len(times)//2]:7.1f} ms")
+    print(f"  max:    {times[-1]:7.1f} ms")
+    print(f"  avg:    {sum(times)/len(times):7.1f} ms")
+
+
+if __name__ == "__main__":
+    bench("PNG", ROOT / "fapiao_1.png", runs=5)
+    bench("PDF", ROOT / "fapiao.pdf", runs=5)

ry-ocr/tests/bench_dpi.py

@@ -0,0 +1,93 @@
+"""基准测试：不同 DPI 下的 PDF 渲染耗时 vs OCR 耗时 vs 识别准确度"""
+import sys
+import time
+from pathlib import Path
+
+try:
+    sys.stdout.reconfigure(encoding="utf-8")
+except Exception:
+    pass
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+import fitz
+from app.core import warmup
+from app.services.recognize_service import recognize_file
+
+
+def render(pdf_path: Path, dpi: int) -> Path:
+    out = ROOT / f"_bench_{dpi}dpi.png"
+    doc = fitz.open(str(pdf_path))
+    page = doc[0]
+    pix = page.get_pixmap(matrix=fitz.Matrix(dpi / 72, dpi / 72), alpha=False)
+    pix.save(str(out))
+    doc.close()
+    return out
+
+
+def bench_dpi(pdf_path: Path, dpi: int, runs: int = 3):
+    img_path = render(pdf_path, dpi)
+    img_bytes = img_path.read_bytes()
+    print(f"\n--- DPI={dpi} | 渲染图 {img_path.stat().st_size//1024}KB ---")
+
+    times = []
+    fields_ok = None
+    for i in range(runs):
+        t0 = time.time()
+        result = recognize_file(img_path.name, img_bytes)
+        elapsed = (time.time() - t0) * 1000
+        times.append(elapsed)
+        if i == 0 and result.success:
+            fields_ok = result.fields
+        print(f"  run {i+1}: {elapsed:7.1f}ms  success={result.success}")
+
+    # 清理中间文件
+    img_path.unlink(missing_ok=True)
+    (ROOT / f".{img_path.stem}_pages").rmdir() if (ROOT / f".{img_path.stem}_pages").exists() else None
+    # 上面可能删不了子目录，try 再清理 _pages 下文件
+    pages_dir = ROOT / f".{img_path.stem}_pages"
+    if pages_dir.exists():
+        for f in pages_dir.glob("*"):
+            f.unlink(missing_ok=True)
+        pages_dir.rmdir()
+
+    times.sort()
+    print(f"  min={times[0]:.1f}ms  median={times[len(times)//2]:.1f}ms  max={times[-1]:.1f}ms")
+
+    return fields_ok
+
+
+if __name__ == "__main__":
+    pdf = ROOT / "fapiao.pdf"
+    print(f"PDF: {pdf.name}, {pdf.stat().st_size//1024}KB, 1 页")
+
+    warmup()
+    print("预热完成\n")
+
+    print("=" * 60)
+    fields_100 = bench_dpi(pdf, 100)
+    print("=" * 60)
+    fields_150 = bench_dpi(pdf, 150)
+    print("=" * 60)
+    fields_200 = bench_dpi(pdf, 200)
+    print("=" * 60)
+    fields_300 = bench_dpi(pdf, 300)
+    print("=" * 60)
+
+    # 汇总
+    print("\n\n=== 关键字段对比 ===")
+    print(f"{'字段':<15} {'100dpi':<25} {'150dpi':<25} {'200dpi':<25} {'300dpi':<25}")
+    key_fields = ["invoice_no", "invoice_date", "amount", "amount_cn",
+                  "amount_pretax", "tax_amount", "seller_name", "buyer_name",
+                  "seller_tax_no", "buyer_tax_no", "amount_match"]
+    for k in key_fields:
+        row = [k]
+        for f in [fields_100, fields_150, fields_200, fields_300]:
+            v = getattr(f, k) if f else None
+            if v is None:
+                row.append("(空)")
+            else:
+                s = str(v)[:23]
+                row.append(s)
+        print(f"{row[0]:<15} {row[1]:<25} {row[2]:<25} {row[3]:<25} {row[4]:<25}")

ry-ocr/tests/test_invoice_extractor.py

@@ -0,0 +1,91 @@
+"""单元测试：发票字段抽取（不依赖 PaddleOCR）
+
+直接喂文本，验证字段抽取与金额校验逻辑。
+"""
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+# 强制 UTF-8 输出（Windows cp936 默认中文乱码）
+try:
+    sys.stdout.reconfigure(encoding="utf-8")
+except Exception:
+    pass
+
+from app.services import extract_invoice  # noqa: E402
+
+
+SAMPLE_VAT_ELEC = """
+增值税电子普通发票
+发票代码: 011002000000
+发票号码: 12345678
+开票日期: 2024年05月20日
+名称: 上海某科技有限公司
+纳税人识别号: 91310115MA1K3X9Y8A
+名称: 北京某某贸易有限公司
+纳税人识别号: 91110108MA01K3X9Y8A
+金额 ¥1234.56
+税额 ¥74.07
+价税合计 ¥1308.63
+（贰仟零捌元陆角叁分）
+备注:
+"""
+
+
+SAMPLE_VAT_SPECIAL = """
+增值税专用发票
+发票代码 011002100111
+发票号码 87654321
+开 票 日 期: 2023年12月01日
+名        称: 深圳某有限公司
+纳税人识别号: 91440300MA5DCBA123
+名        称: 广州某科技股份公司
+纳税人识别号: 91440101MA59ABC987
+不含税价 ¥10000.00   税率 13%
+税    额  ¥1300.00
+价税合计（大写）壹万壹仟叁佰元整 （小写）¥11300.00
+"""
+
+
+def test_vat_electronic():
+    fields = extract_invoice(SAMPLE_VAT_ELEC)
+    print(fields.model_dump_json(indent=2))
+    assert fields.invoice_type == "增值税电子普通发票"
+    assert fields.invoice_code == "011002000000"
+    assert fields.invoice_no == "12345678"
+    assert fields.invoice_date == "2024-05-20"
+    assert fields.amount is not None and abs(fields.amount - 1308.63) < 0.01
+
+
+def test_vat_special():
+    fields = extract_invoice(SAMPLE_VAT_SPECIAL)
+    print(fields.model_dump_json(indent=2))
+    assert fields.invoice_type == "增值税专用发票"
+    assert fields.invoice_no == "87654321"
+    assert fields.invoice_date == "2023-12-01"
+    assert fields.amount is not None and abs(fields.amount - 11300.0) < 0.01
+    # 大写金额提取（壹万壹仟叁佰元整）
+    assert fields.amount_cn is not None
+    assert "壹万" in fields.amount_cn or "万" in fields.amount_cn
+    # 大写 vs 小写一致
+    assert fields.amount_match is True
+
+
+def test_amount_consistency():
+    """大写小写一致"""
+    from app.utils import amount_consistent
+    assert amount_consistent("贰佰元整", 200.0) is True
+    assert amount_consistent("壹仟元整", 1000.0) is True
+    # 不一致
+    assert amount_consistent("贰佰元整", 300.0) is False
+
+
+if __name__ == "__main__":
+    test_vat_electronic()
+    print("-" * 60)
+    test_vat_special()
+    print("-" * 60)
+    test_amount_consistency()
+    print("✅ all tests passed")