Adding a lower-level event and a logger example

[datvo06]

Fix #381

[jfeser]

This looks mostly good! Two comments:

1. llm_request should have essentially the same interface as client.responses.create:

@defop
def llm_request(client: openai.OpenAI, *args, **kwargs) -> Any:
    """Low-level LLM request. Handlers may log/modify requests and delegate via fwd()."""
    return client.responses.create(*args, **kwargs)

2. The notebook needs to be rerun since it has an exception.

[jfeser]

You don't have to add it in this PR, but I'd be happy to have a llm logging handler that exposes a python logger and logs llm_request and tool_call.

[datvo06]

You don't have to add it in this PR, but I'd be happy to have a llm logging handler that exposes a python logger and logs llm_request and tool_call.

I see, that makes sense, I can make a short attempt, if it is too hard, then I'll revert.

[datvo06]

I added the logger that expose Python logger:

# 1. Create a logger
logger = logging.getLogger("effectful.llm")
logger.setLevel(logging.INFO)
h = logging.StreamHandler(sys.stdout)
h.setLevel(logging.INFO)
h.setFormatter(logging.Formatter("%(levelname)s %(name)s %(message)s %(payload)s"))
logger.addHandler(h)

# 2. Pass it to the handler
llm_logger = LLMLoggingHandler(logger=logger)

# Avoid cache for demonstration
try:
    haiku.cache_clear()
except Exception:
    pass

with handler(provider), handler(llm_logger):
    _ = haiku("fish3")
    _ = limerick("fish4")

[jfeser]

This shouldn't be a configuration parameter. Instead the logger should be exposed as an attribute.

[jfeser]

Still outstanding.

[jfeser]

Still outstanding.

[jfeser]

Happy to merge once you address outstanding comments.

[datvo06]

Updated: only have one optional parameter logger in LLMLoggingHandler.