Split up RequestFormatter

[sjmonson]

Summary

Splits the RequestFormatter into multiple separate components across data loading, data preprocessing, and the backend. Necessary step for next attempt at multi-turn.

Details

• Removed the RequestFormatter preprocessor
• Added "Encode" preprocessors for audio, video, and image
• Added a "Finalizer" stage after preprocessors to convert from column ordered data to request object
• Added most of previous formatter code to backend

Test Plan

• Run all know benchmark well-lit pathways

Related Issues

• Resolves #

---

• "I certify that all code in this PR is my own, except as noted below."

Use of AI

• Includes AI-assisted code completion
• Includes code generated by an AI application
• Includes AI-generated tests (NOTE: AI written tests should have a docstring that includes ## WRITTEN BY AI ##)

[sjmonson]

A note on --request-type: Request type only really makes sense in the context of the openai_http backend. For upcoming backends (such as #178 and #548) I imagine request_type will be completely ignored. Ideally we would not give openai_http any special treatment and require request_type to either be specified as a --backend-kwarg (basically how its implemented now without the special --request-type shorthand) or as entirely separate backend for each request-type. For now I have kept --request-type to avoid user confusion.