feat: refactor summarizer and PDF extraction pipeline

- Split summarizer into summary_generator and summary_persister modules
- Refactor pdf_image_extractor to two-phase pipeline with PicoDet layout detection
- Add layout_detector service for PicoDet-S_layout_3cls integration
- Add exceptions module with ConflictError and NotFoundError
- Improve admin dashboard with better statistics and task management
- Add design review document with system optimization suggestions
- Add new tests for crawler, pdf_downloader, pipeline, and summary_utils
- Update dependencies and configuration
- Clean up dead code and improve error handling

app/services/pi_client.py

@@ -11,6 +11,7 @@ import uuid
 from pathlib import Path
 
 from app.config import settings
+from app.utils import truncate_error
 from app.services.summary_utils import (
     JsonNotFoundError,
     build_prompt,
@@ -21,6 +22,9 @@ from app.services.summary_utils import (
 
 logger = logging.getLogger(__name__)
 
+# PDF 全文注入模式的字符上限 — 超过此阈值自动切换到 search 模式
+_PDF_MAX_CHARS = 80_000
+
 # 重新导出，保持向后兼容
 __all__ = [
     "PiTimeoutError",
@@ -45,7 +49,7 @@ class PiProcessError(Exception):
     def __init__(self, returncode: int, stderr: str):
         self.returncode = returncode
         self.stderr = stderr
-        super().__init__(f"pi exited with code {returncode}: {stderr[:500]}")
+        super().__init__(f"pi exited with code {returncode}: {truncate_error(stderr)}")
 
 
 # ── pi CLI 调用 ────────────────────────────────────────────────────────
@@ -72,23 +76,27 @@ async def call_pi(
 
     actual_mode = pdf_mode
     if pdf_mode == "auto":
-        if txt_size > 80_000:
+        if txt_size > _PDF_MAX_CHARS:
             actual_mode = "search"
             logger.info(
-                "Auto mode: %s text=%d chars > 80k → search", arxiv_id, txt_size
+                "Auto mode: %s text=%d chars > %dk → search",
+                arxiv_id, txt_size, _PDF_MAX_CHARS // 1000,
             )
         else:
             actual_mode = "inject"
             logger.info(
-                "Auto mode: %s text=%d chars ≤ 80k → inject", arxiv_id, txt_size
+                "Auto mode: %s text=%d chars ≤ %dk → inject",
+                arxiv_id, txt_size, _PDF_MAX_CHARS // 1000,
             )
 
     # inject 模式需要截断过长的文本（避免撑爆 context）
-    if actual_mode == "inject" and txt_size > 80_000:
+    if actual_mode == "inject" and txt_size > _PDF_MAX_CHARS:
         body = txt_path.read_text(encoding="utf-8")
-        trimmed = body[:80_000].rstrip()
+        trimmed = body[:_PDF_MAX_CHARS].rstrip()
         txt_path.write_text(trimmed, encoding="utf-8")
-        logger.info("Truncated %s for inject: %d → %d chars", arxiv_id, txt_size, len(trimmed))
+        logger.info(
+            "Truncated %s for inject: %d → %d chars", arxiv_id, txt_size, len(trimmed)
+        )
 
     prompt_text = build_prompt(arxiv_id, meta_path, txt_path, actual_mode, fix_errors)
 
@@ -101,7 +109,8 @@ async def call_pi(
     cmd = [
         settings.PI_BIN,
         "-p",
-        "--tools", tools,
+        "--tools",
+        tools,
     ]
     if fix_errors:
         cmd += ["--session", session_id, "--continue"]
@@ -118,10 +127,14 @@ async def call_pi(
 
     logger.info(
         "Calling pi for %s (fix=%s, session=%s, mode=%s)",
-        arxiv_id, bool(fix_errors), session_id, actual_mode,
+        arxiv_id,
+        bool(fix_errors),
+        session_id,
+        actual_mode,
     )
 
     import time as _time
+
     _t_sub_start = _time.monotonic()
 
     proc = await asyncio.create_subprocess_exec(
@@ -151,7 +164,9 @@ async def call_pi(
 
     logger.info(
         "pi subprocess for %s: %.2fs%s",
-        arxiv_id, _t_sub_end - _t_sub_start, _file_info,
+        arxiv_id,
+        _t_sub_end - _t_sub_start,
+        _file_info,
     )
 
     if proc.returncode != 0: