Remove LoggingLLMHandler

[kiranandcode]

This PR adds documentation for the fact that weave.op is compatible with effectful and litellm.completions. In particular to get weave logging for template calls it simply suffices to add the @weave.op decorator to a template definition.

@weave.op
@Template.define
def haiku(topic: str) -> str:
  """Generate a haiku about {topic}."""
   raise NotHandled

Closes #493

[kiranandcode]

The issue with this PR is that, as @weave.op wraps the template definition it no longer gets picked up as a tool for the LLM to use.

@weave.op
@Template.define
def haiku(topic: str) -> str:
    """Generate a haiku about the topic {topic}."""
    raise NotImplementedError

@weave.op
@Template.define
def poetry_compilation(topic: str) -> list[str]:
    """Generate a collection of poems surrounding {topic}, calling `haiku` on related topics to generate the elements."""
    raise NotImplementedError


with handler(LiteLLMProvider()):
    for poem in poetry_compilation('apple'):
        print(poem)
        print('')

[kiranandcode]

Following discussions from #509, updating documentation to instead describe adding a custom handler.

[eb8680]

Closing #493 would require deleting the current logging handler from providers.py. Also, I'm just not sure we want to add weave as a library-level dependency in effectful[llm] or include this particular code snippet in the documentation when it's not something we've tested extensively ourselves - I can believe this code runs, but it's not associated with any unit tests and there's no way for someone to tell from this notebook cell if it solves the right problem and doesn't interact poorly with other components in surprising ways. I really think we should just leave out any recommendation like this unless it's something we've tested extensively ourselves or it's something built into LiteLLM.

[kiranandcode]

okay sounds good. I'll drop the documentation change and drop the logging handler.

[kiranandcode]

@eb8680 dropped llmlogginghandler. Yixiu makes a lot of use of wandb, so I'll ask him to try out the wandb handler just in MARA for now and if it's popular and gets used internally sufficiently then we can document it later.