feat: 新增 .txt 纯文本文件支持

- 添加 txt 到允许上传格式，跳过 MinerU OCR
- 直接读取 UTF-8 文本，按 3000 字符/页分页后送入 LangExtract
- 适合有中文文本的规范文件，避免扫描件 OCR 乱码问题

backend/routers/system.py

@@ -110,6 +110,7 @@ async def list_formats():
             {"ext": "jpg",  "description": "JPEG 图片（单页）", "max_size_mb": 200, "max_pages": 1, "requires_ocr": True},
             {"ext": "jpeg", "description": "JPEG 图片（单页）", "max_size_mb": 200, "max_pages": 1, "requires_ocr": True},
             {"ext": "html", "description": "HTML 文件", "max_size_mb": 200, "max_pages": 600, "requires_ocr": False},
+            {"ext": "txt",  "description": "纯文本文件（UTF-8 编码）", "max_size_mb": 200, "max_pages": 0, "requires_ocr": False},
         ],
         "ocr_languages": [
             {"code": "ch", "name": "中文（默认）"},

backend/services/document_service.py

@@ -7,7 +7,7 @@ from pathlib import Path
 
 from storage import file_store as fs
 
-ALLOWED_EXTENSIONS = {"pdf", "docx", "doc", "pptx", "ppt", "png", "jpg", "jpeg", "html"}
+ALLOWED_EXTENSIONS = {"pdf", "docx", "doc", "pptx", "ppt", "png", "jpg", "jpeg", "html", "txt"}
 MAX_FILE_SIZE_MB = 200
 
 

backend/services/indexing_service.py

@@ -80,6 +80,50 @@ def _run_pipeline(job_id: str) -> None:
             _update_meta(job_id, status="cancelled", stage="Cancelled")
             return
 
+        from pipeline.text_assembler import PageText, BlockSpan
+        from pipeline.entity_extractor import create_model, extract_entities
+        from pipeline.kg_builder import build_kg, extractions_to_records
+
+        ext = pdf_path.suffix.lower().lstrip(".")
+
+        if ext in ("txt",):
+            # Plain text — skip MinerU, split by paragraphs
+            _update_meta(job_id, status="parsing", stage="Parsing plain text...")
+            raw = pdf_path.read_text(encoding="utf-8")
+
+            # Split into ~3000-char pages at double-newline boundaries
+            PAGE_SIZE = 3000
+            raw_pages = []
+            pos = 0
+            while pos < len(raw):
+                end = min(pos + PAGE_SIZE, len(raw))
+                if end < len(raw):
+                    # Try to break at a double newline
+                    br = raw.rfind("\n\n", pos, end)
+                    if br > pos + PAGE_SIZE // 2:
+                        end = br + 2
+                raw_pages.append(raw[pos:end])
+                pos = end
+
+            pages = []
+            for i, text in enumerate(raw_pages):
+                pages.append(PageText(
+                    page_idx=i,
+                    text=text.strip(),
+                    block_spans=[BlockSpan(
+                        block_index=0, block_type="text",
+                        page_idx=i, char_start=0,
+                        char_end=len(text.strip()), bbox=[0, 0, 0, 0],
+                    )],
+                ))
+            total_pages = len(pages)
+            block_types = {"text": total_pages}
+
+            # Save full.txt in mineru_output for reference
+            mineru_out_dir = job_dir / "mineru_output"
+            mineru_out_dir.mkdir(parents=True, exist_ok=True)
+            (mineru_out_dir / "full.md").write_text(raw, encoding="utf-8")
+        else:
             _update_meta(job_id, status="parsing", stage="MinerU document parsing...")
             mineru_out_dir = job_dir / "mineru_output"
             mineru_out_dir.mkdir(parents=True, exist_ok=True)
@@ -96,7 +140,8 @@ def _run_pipeline(job_id: str) -> None:
                 raise RuntimeError(f"MinerU failed: {result.stderr[:500]}")
 
             # Find content_list.json in MinerU output
-        # MinerU writes output to mineru_mvp/output/{stem}/
+            from pipeline.text_assembler import load_content_list, assemble_pages, count_blocks_by_type
+
             stem = pdf_path.stem
             mineru_default_out = MINERU_PIPELINE.parent / "output" / stem
             content_list_path = None
@@ -105,12 +150,10 @@ def _run_pipeline(job_id: str) -> None:
                 matches = list(mineru_default_out.glob("*_content_list.json"))
                 if matches:
                     content_list_path = matches[0]
-                # Copy to our job dir
                     import shutil
                     shutil.copytree(str(mineru_default_out), str(mineru_out_dir), dirs_exist_ok=True)
 
             if not content_list_path:
-            # Fallback: search job mineru_output dir
                 matches = list(mineru_out_dir.glob("*_content_list.json"))
                 if matches:
                     content_list_path = matches[0]
@@ -118,15 +161,6 @@ def _run_pipeline(job_id: str) -> None:
             if not content_list_path or not content_list_path.exists():
                 raise RuntimeError(f"MinerU output content_list.json not found. stdout: {result.stdout[:300]}")
 
-        # ── Stage 2: extracting ───────────────────────────────────────────
-        if _cancel_flags.get(job_id):
-            _update_meta(job_id, status="cancelled", stage="Cancelled")
-            return
-
-        from pipeline.text_assembler import load_content_list, assemble_pages, count_blocks_by_type
-        from pipeline.entity_extractor import create_model, extract_entities
-        from pipeline.kg_builder import build_kg, extractions_to_records
-
             content_list = load_content_list(content_list_path)
             pages = assemble_pages(content_list)
             total_pages = len(pages)