fix error when deploying a model from mlflow

[Rusteam]

Add config_path key to copy_paths to avoid KeyError when deploying from mlflow models in onnx flavor with config.pbtxt

[GuanLuo]

Hi @Rusteam, thanks for the contribution. The change looks good but do you mind to sign the CLA as instructed here before we can merge this pull request?

[Rusteam]

I don't mind. But I'm not a corporation, I'm an individual contributor. Is it fine?

[Rusteam]

@GuanLuo I've sent a signed CLA to the email mentioned. Can you review and merge now?

I've also added two other commits: to fix ensemble deployment and labels.txt if present

[GuanLuo]

@Rusteam sorry that I missed the earlier message, we have received your CLA and the change looks good to me. I would like to extend L0_mlflow test to cover this case where ONNX model is logged with additional files (i.e. config / labels). I wonder if you have any insight on logging ONNX flavor in this form, currently the test is using the MLFlow ONNX plugin to log the model, but I don't think mflow.onnx.log_model can accept additional files

[Rusteam]

You're right it does not. What I do is I log other files as artifacts right after logging the model, i.e.:

with mlflow.start_run():
    mlflow.onnx.log_model(model, path)
    mlflow.log_artifact("labels.txt", path)
    mlflow.log_artifact("config.pbtxt", path)

[GuanLuo]

Thanks for the hint, I have extended the test to cover the case with config in this PR #4469, feel free to leave any comments