Clarify the semantic conventions of AI spans. #4865
Description
This document from Semantic conventions for generative client AI spans uses the "gen_ai.request.presence_penalty" field, while the code uses the "llm.presence_penalty" field. Which one should be the standard?
There are other fields with the same issue ("gen_ai.request.stop_sequences","gen_ai.request.frequency_penalty", "gen_ai.response.finish_reasons", "gen_ai.request.top_k" )
Docs: https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-spans/