feat: implement PP-Structure table extraction pipeline with GPU runtime configuration support

src/ocr_sprint/api/routes/documents.py

@@ -28,13 +28,6 @@ from functools import partial
 from typing import Annotated
 from uuid import UUID, uuid4
 
-# Thread pool dedicated to blocking OCR work. Using a *separate* pool
-# (rather than the default loop executor) lets us cap the number of
-# concurrent heavy OCR jobs independently of other thread-pool users.
-# With 1 Celery worker + 1 sync slot we never exceed 2 parallel OCR
-# runs; keep the pool at 1 so RAM stays bounded on the 7.4 GB server.
-_OCR_EXECUTOR = ThreadPoolExecutor(max_workers=1, thread_name_prefix="ocr-inline")
-
 from fastapi import (
     APIRouter,
     Depends,
@@ -73,6 +66,13 @@ from ocr_sprint.schemas.review import (
 from ocr_sprint.storage.blob import get_blob_storage
 from ocr_sprint.utils.logging import get_logger
 
+# Thread pool dedicated to blocking OCR work. Using a *separate* pool
+# (rather than the default loop executor) lets us cap the number of
+# concurrent heavy OCR jobs independently of other thread-pool users.
+# With 1 Celery worker + 1 sync slot we never exceed 2 parallel OCR
+# runs; keep the pool at 1 so RAM stays bounded on the 7.4 GB server.
+_OCR_EXECUTOR = ThreadPoolExecutor(max_workers=1, thread_name_prefix="ocr-inline")
+
 router = APIRouter(
     prefix="/documents",
     tags=["documents"],
@@ -99,18 +99,17 @@ def _row_to_response(row: object) -> DocumentResponse:
 
     assert isinstance(row, JobRow)
     status_enum = DocumentStatus(row.status)
-    personel_list = None
+    result_obj = None
     if row.result is not None:
         result_obj = ExtractionResult.model_validate(row.result)
         # Auto-number personnel entries sequentially (1, 2, 3, ...)
         for idx, entry in enumerate(result_obj.personel, start=1):
             entry.no = idx
-        personel_list = result_obj.personel
     return DocumentResponse(
         job_id=row.job_id,
         status=status_enum,
         confidence=row.confidence,
-        data=personel_list,
+        data=result_obj,
         review_flags=list(row.review_flags or []),
         error=row.error,
         approved=bool(row.approved),

src/ocr_sprint/api/routes/health.py

@@ -6,6 +6,7 @@ from fastapi import APIRouter
 from fastapi.responses import JSONResponse
 
 from ocr_sprint import __version__
+from ocr_sprint.config import get_settings
 from ocr_sprint.pipeline import ocr as _ocr
 from ocr_sprint.pipeline import table as _table
 
@@ -21,15 +22,18 @@ async def health() -> dict[str, str]:
 @router.get("/health/ready")
 async def readiness() -> JSONResponse:
     """Readiness check — returns 200 when OCR models are loaded, 503 if still warming up."""
+    settings = get_settings()
     ocr_ready = _ocr._instance is not None
-    table_ready = _table._instance is not None
+    table_ready = (not settings.tables_enabled) or _table._instance is not None
     ready = ocr_ready and table_ready
     payload = {
         "status": "ready" if ready else "warming_up",
         "version": __version__,
         "models": {
             "paddleocr": "ready" if ocr_ready else "loading",
-            "pp_structure": "ready" if table_ready else "loading",
+            "pp_structure": (
+                "disabled" if not settings.tables_enabled else "ready" if table_ready else "loading"
+            ),
         },
     }
     return JSONResponse(content=payload, status_code=200 if ready else 503)