If there is an error raised during a AI span the status of the span should follow the OpenTelementry definition of a span status: https://opentelemetry.io/docs/concepts/signals/traces/#span-status

For AI spans, if there is an error we must set the status to 'error', otherwise we do not set the status and let relay set the default status. (reasoning: If the AI span completes without error, it could also be that the LLM answered with "I can not do this" which basically means the span was not successful. It is hard to detect if a LLM request was successful or not, so we keep the status unset)