fix(ocr): reduce memory usage for 16GB dev machines

- Surya models lazy-load on first OCR request instead of at startup
  (saves ~3-4GB idle RAM — Kraken stays eager at ~16MB)
- Process one page at a time in Surya engine (limits peak memory)
- RECOGNITION_BATCH_SIZE=1, DETECTOR_BATCH_SIZE=1 (slower but fits in RAM)
- Revert mem_limit back to 6GB (sufficient with these optimizations)
- Render DPI stays at 200

Idle memory: ~2GB (Kraken only). Peak during OCR: ~5-6GB (Surya loaded).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ocr-service/main.py

@@ -22,14 +22,13 @@ _models_ready = False
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    """Load all OCR models at startup before accepting requests."""
+    """Load lightweight models at startup. Surya loads lazily on first request."""
     global _models_ready
 
-    logger.info("Loading OCR models at startup...")
-    surya_engine.load_models()
+    logger.info("Loading Kraken model at startup (Surya loads lazily on first OCR request)...")
     kraken_engine.load_models()
     _models_ready = True
-    logger.info("All OCR models loaded — ready to accept requests")
+    logger.info("Startup complete — ready to accept requests")
 
     yield