Add OpenAIBatch backend and refactor RequestProcessor to be compatible

[RyanMarten]

Now you can use batch via

curator.Prompter(
    prompt_func=prompt_func, 
    parse_func=parse_func, 
    model_name="gpt-4o-mini", 
    batch=True
)

New example in examples/distill.py which shows this for a large reannotation job.

Also includes small change to the interface:
prompt_func returns a dictionary
{ "user_prompt": content, "system_prompt": content}
or
{ "user_prompt": content: "system_prompt": content}
to
prompt_func returns a list (of standard form messages or string
content
which gets converted to

[
 {"role": "user", "content": content}
]

or the messages

[
 {"role": "system", "content": content}
 {"role": "user", "content": content},
]

[vutrung96]

LGTM