Extend 'create_judge' command to handle multimodal judges

[rchan26]

Currently, the create_judge script only works with evaluating standard LLM outputs taking in the textual prompt and textual response. For evaluating multimodal LLMs, we may need to input media inputted with the prompt into the judge too. Potential easy fix is to add the multimodal field that is in the prompt to the judge evaluation if it exists.