fix(langchain): don't record tool call output if not include_prompt / should_send_default_pii

[shellmayr]

• This PR makes sure we don't record tool calls as outputs in gen_ai.chat spans in langchain when users deselect the option to include prompts / send pii

Closes TET-1181

[linear]

TET-1179 Langgraph missing execute tool spans

[codecov]

✅ All tests passed in 2567.23s

[linear]

TET-1181 Langchain missing input output on execute_tool span

[alexander-alderman-webb]

My one concern is that I have not heard tool arguments referred to as "prompts". Would users be confused if setting include_prompts=False also hides data from the tool calls?

We could consider more granular setting, especially since I am unsure how intuitive it is for tool call input and outputs to fall under the "prompt" umbrella.

But if that's more niche than we want to cater for, this looks good to go.

[shellmayr]

My one concern is that I have not heard tool arguments referred to as "prompts". Would users be confused if setting include_prompts=False also hides data from the tool calls?

@alexander-alderman-webb My thought on this is that a tool call may include raw user input, and that raw user input is a prompt, no? And in terms of behaviour, if a user specifies that they do not want to record that type of user input in general, we should stick to it throughout the SDK. Just to be clear: we still record 1. that tool calls happen and 2. which tool is being called - just not the user input associated with it.

[alexander-alderman-webb]

Looks good to me. Thanks for answering my questions!

Not sure if someone with more experience on the AI integrations should also approve.

[antonpirker]

Yes, looks good. The name include_prompts is not the best, you are right @alexander-alderman-webb but this is the name we have in other integrations. The functionality of it evolved over time. We probably should think of changing it but as this is a library we need to be careful with this to not break users code.