Add support for creating inference-only SageMaker pipelines

Currently inference tasks in our SageMaker pipeline generator always get their model input path by a variable [model_output_path](https://github.com/thvasilo/graphstorm/blob/2928c40553477fd5ab330a70bda859a28443a605/sagemaker/pipeline/create_sm_pipeline.py#L574) that gets set during the creation of a training task.

We should add an optional CLI argument that allows users to provide their own path for model artifacts, which will make it possible to define inference-only pipelines with job sequences like `gconstruct inference` or `gsprocessing dist_part inference`.