[FEATURE] Support non-chain-style cocoindex function call

[badmonster0]

Currently cocoindex function can only be called in chain style (arg0.transform(FunctionSpec(...), remaining_args...)). It's convenient for many cases, but there're situations that not fits into this. e.g. in Python we may have a function like:

def extract_from_llm(text: str = None, image: bytes = None):
  ...

If the first argument text is required, text can acts as a self (this is the current chain-style transform that we support). But here both text and image are optional. There's no clear equivalence in chain style.

For the non-chain-style function calls, two options in my mind for now:

1. FunctionSpec(...).transform(args...) or FunctionSpec(...).call(args...). For standalone function, it's custom_fn.transform(args...) or custom_fn.transform.call(args...)

   • More explicit
   • But more verbose, and FunctionSpec cannot have call field

2. FunctionSpec(...)(args...). For standalone function, it's custom_fn(args...)

   • More concise, no restriction
   • But two adjacent (...)(...) looks a little bit uncommon
   • For standalone function, it's probably hard to avoid mypy raising error (mypy will assume custom_fn takes values as input, instead of DataSlice)

Would like to hear some feedback about the desired style before proceeds.

More context: #653 (comment)

---

❤️ Contributors, please refer to 📙Contributing Guide.
Unless the PR can be sent immediately (e.g. just a few lines of code), we recommend you to leave a comment on the issue like I'm working on it or Can I work on this issue? to avoid duplicating work. Our Discord server is always open and friendly.

[lemorage]

Cool! IMO, the second option seems more intuitive to me and quite pythonic in some sense. There are no restrictions here, and easy to deal with. Take the image_search as an example:

    with data_scope["images"].row() as img:
        extract_func = cocoindex.functions.ExtractByLlm(
            llm_spec=cocoindex.llm.LlmSpec(
                api_type=cocoindex.LlmApiType.OPENAI,
                model="gpt-4o",
            ),
            instruction="Describe the image in one detailed sentence. ",
            output_type=str,
        )
        caption_ds = extract_func(image=img["content"])  # Non-chain-style
        # caption_ds = img["content"].transform(extract_func) # Chain-style

I can take this if the plan is settled, WDYT?

[badmonster0]

Sounds good! Option 2 is also what I prefer! Assigned.