FEAT support langsmith for tracing

[leonj1]

Fixes #9

Since this is a langgraph project, then it's possible to use langsmith to see what's happening.
I do not know how to use langsmith (yet), but there is a connection now if needed for debugging.

[ai-christianson]

Looks like some conflicts (hopefully minor) after merging the logging PR.

I think the feature idea itself makes a ton of sense because right now we have zero visibility into the language model calls themselves.

Basically has been manual prompt tuning and making sure is works in real world scenarios.

This feature will give us visibility into the LLM calls which will be key for the bigger picture.

I'm going to put another ticket for benchmarking on SWE bench. Between langsmith visibility, logging, and benchmarks, we'll have a much more solid foundation for making sustainable progress.

[leonj1]

I'm unsure if I should proceed with PR since I don't know if langgraph is meant to integrate with langsmith. I have found no article that gives me confidence in this intent. I'd rather not force an integration that's not meant to be.

I fixed the conflicts if you still think it's reasonable to merge this.

[ai-christianson]

Yeah that's fair.

For now we can leave this issue open and treat it more like an exploration/research thread and figure out if it's something worth fully integrating.

Edit: but the general idea of tracing and visibility is a good one. It can get to be fairly opaque what is actually going on under the hood with all the prompting.

The visibility would also be instrumental in optimizing the agent for token use as well as making it work better with other models.