support using inference_base_image in transformers processor

[happywhaler]

azureml-assets/assets/training/model_management/src/azureml/model/mgmt/processors/transformers/convertors.py

Line 183 in 69c9dff

metadata["azureml.base_image"] = "mcr.microsoft.com/azureml/curated/foundation-model-inference:" \

        if self._inference_base_image:
            metadata["azureml.base_image"] = self._inference_base_image
        elif self._vllm_enabled:
            mlclient = get_mlclient("azureml")
            vllm_image = mlclient.environments.get("foundation-model-inference", label="latest")
            metadata["azureml.base_image"] = "mcr.microsoft.com/azureml/curated/foundation-model-inference:" + str(
                vllm_image.version
            )
            logger.info("Metadata: {}".format(metadata))

we don't always want to take the latest vllm image, would be awesome to be able to use this argument here as well as latest is not always working (alternatively, allow setting a different label).