This is more or less working but has some todos for fixing the aggregation on mt_mbpp.

Thanks@IanMagnusson ! do u mind:

• Moving these into a subdir likescripts/visualize_baselines_on_benchmark/ with aREADME.md for how to invoke? For README, the main questions:
  ** I have is what aspects of each script I need to modify for a given new benchmark?
  ** When do I need to usedownload_checkpoints.sh? Was this something you already ran one time? What$OUTPUT_DIR did you use?
  ** Doesbaseline_task_viz.py consumebaseline_sampler_data.json? If so, where did you getbaseline_sampler_data.json in first place?
• Migratebaseline_sampler_data.json which is currently in root also within this subdir