fix: change default OCR language from English to Swedish

Project targets Swedish invoice extraction. PaddleOCR sv model provides
better recognition of Swedish-specific characters (å, ä, ö).

packages/training/training/processing/gpu_pool.py

@@ -48,7 +48,7 @@ def _init_gpu_worker(gpu_id: int = 0) -> None:
         from paddleocr import PaddleOCR
 
         # PaddleOCR 3.x init - minimal params, GPU controlled via paddle.set_device
-        _ocr_instance = PaddleOCR(lang="en")
+        _ocr_instance = PaddleOCR(lang="sv")
         _gpu_initialized = True
         logger.info(f"GPU worker initialized on GPU {gpu_id} in process {os.getpid()}")