Suggestion: Refactor so that batch LLM calls are default without needing to use async wrapper

[bendavidsteel]

I wanted to use batch LLM calls, realised that I'd need to use an async wrapper to do that, and I found that makes the naming somewhat unclear. So maybe all LLM calls should be batch by default and let the call implementation handle lists with one element?

[lmcinnes]

I think this is probably the right call. We started with the simplest approach, and worked up from there. Once we have more testing and use of the Async wrappers to be sure they are all working well I agree that it will make sense to swap things around, and have a "Serial" wrapper instead.