Enable input column, label column for vision tasks

[rdondera-microsoft]

Context (benchmarking the Stable Diffusion model on MSCOCO): main...rdondera/sd

Honor column name parameters (inputs, label) passed to the model prediction component for vision tasks. This does not bring benefits for image classification, object detection or instance segmentation, but is needed for image generation tasks, where the input name depends on the model.

Also: clarified how vision dataset processing works in the model evaluation pipeline. The MLTable is assumed to have columns "image_url" and "label" and the Pandas dataframe created from it has columns set according to the input_column_names and label_column_name parameters. When these parameters are not specified, task specific default values are filled in.

Also: brought back changes from #2888 , which were removed in a subsequent PR.

Tested by running the evaluation pipeline for ViT in the cloud: https://ml.azure.com/experiments/id/9f7237b4-3232-49d3-ae02-e07326242835?wsid=/subscriptions/dbd697c3-ef40-488f-83e6-5ad4dfb78f9b/resourcegroups/rdondera/providers/Microsoft.MachineLearningServices/workspaces/benchmarking&tid=72f988bf-86f1-41af-91ab-2d7cd011db47 , vit_*.

1. vit_params: both input_column_names and label_column_name properly set ("image", "label")
2. vit_params: neither input_column_names nor label_column_name set
3. vit_wrong_params: both input_column_names and label_column_name set to wrong values ("rgb_image", "class")
4. vit_wrong_label: input_column_names properly set, label_column_name set to wrong value ("image", "class,attribute")

[github-actions]

Test Results for training-model-mgmt-unittests

44 tests   44 ✅  8s ⏱️
 1 suites   0 💤
 1 files     0 ❌

Results for commit 9bb6ea7.