More Granular Target Objects for Langfuse Evaluators

[qrosseel]

Describe the feature or potential improvement

I can only select Trace and Dataset when I select the scope of a Langfuse-hosted evaluation.

Given that traces sometimes contain multiple LLM generations it would be of great value if I can scope my evaluation more granularly.

Benefits:

• Cleaner evaluation prompts (I don't need to tell it to ignore irrelevant Trace IO)
• Less wasted input tokens

So instead of only being able to target Trace, it would be nice if I could either pattern-match or select Generation as the target object.

Additional information

No response