Refactor adding of Langfuse callbacks in PMAT, PMA, Prophet

[kongzii]

Check out this comment https://github.com/gnosis/prediction-market-agent-tooling/pull/345/files#r1718271854

How could it be done nicer? Some ideas:

1. Refactor these lines into a separate function, so it would be like

config = prepare_config_with_langfuse(existing_config, add_langfuse_callback=add_langfuse_callback)

Where existing_config could be just None.

2. Accept callbacks in function arguments, and always create a second function that will pass the callback, if add_langfuse_callback is True:

def infer_category(..., callbacks=None): ...

def infer_category_observed(...):
  callbacks = [init langfuse here]
  return infer_category(..., callbacks)

3. Monkey-patch all Langchain invoke methods, automatically adding the langfuse callback. But this would be nicer as an experimental feature in Langfuse itself.

4. Create a wrapper class around Langchain classes that we can import from PMAT, and it will include the callback automatically.

5. ???

[kongzii]

I asked them about (3) https://github.com/orgs/langfuse/discussions/2949

[kongzii]

I think this can be closed, https://github.com/orgs/langfuse/discussions/2949 isn't getting much traction and the current solution of using get_langfuse_config function + observing everything by default, because it can be (1) easily disabled and (2) un-observed version of underlying function can be accessed by __wrapped__ attribute.