Extended experiment configuration backup

[MaramHasanain]

As part of the output of a benchmarking experiment, we should write the full configuration used to file. This can be useful to version experiments (for reproducibility). Example configs to maintain:

• Running time and date of the experiment
• Train and test dataset file names and the name of dataset script used to load them.
• Task script name
• Model
• Base Prompt used
• Learning setup (0-shot, few-shot, etc) and details about the setup (e.g., method used to select few shots)

"Might save all of this as a pickled object"