Properly integrate OpenAI's batch api into all generation scripts

[Kipok]

We already support batch api directly in the model, but this needs to be exposed in all relevant scripts. The main problem API-wise is how do we wait for the results such that we can easily chain scripts together and make sure non-batch and batch api has consistent interface. Maybe write the id request into the jsonl file and then add checks in all relevant places and fail if it's not done?

Here is an initial attempt based on the prior code version, but we need to upgrade it to work with latest main #79