Allow Explicit is_rec Parameter in __call__ Method

[wjh2012]

Thank you for your excellent project. I truly appreciate and respect your work!

While using RapidOCR with my custom Paddle model, I encountered an issue where the character set was not being read correctly. After some investigation, I discovered that during the ONNX conversion process, the ONNX metadata fails to properly load the dictionary. It appears that this behavior is linked to the way the model type (recognition model) is implicitly determined solely based on the file name.

Currently, the call method in the PaddleOCRModelConvert class determines whether the model is for recognition (is_rec) by checking if the model file name contains the substring "rec". This approach relies exclusively on the file naming convention, which may not always accurately reflect the model type.

To address this, I propose adding an explicit is_rec parameter to the call method rather than relying only on file name inference. This change could also resolve the ONNX metadata loading issue for custom Paddle models in RapidOCR.

Below is a code snippet illustrating the proposed modification:

class PaddleOCRModelConvert:
    def __init__(self) -> None:
        self.opset = 14

    def __call__(
        self,
        model_path: InputType,
        save_dir: InputType,
        txt_path: Optional[str] = None,
        is_del_raw: bool = False,
        is_rec: Optional[bool] = None  # New optional parameter
    ) -> str:
        model_name = Path(model_path).stem

        # Use provided is_rec if available; otherwise, determine by model name
        if is_rec is None:
            is_rec = "rec" in model_name
        ...

[SWHL]

Thanks.
Would you like to submit a PR to fix this?