Integrate Langfuse tracing and tag LLM prompt types

[jooha-yoo]

This connects our system to Langfuse now

• calls the configured API keys (added to .env)
• adds suggestion_type metadata to Langfuse traces to distinguish different LLM prompt types, enabling evaluators to run only on relevant generations and improving clarity when analyzing/comparing prompt evaluations.

This is a draft PR because although tracing is fully connected, LLM-as-judge evaluators are still being configured in the Langfuse UI itself. Evaluation criteria/measures are identified and will be added shortly.