Panel on neural relational data: Tabular foundation models, LLMs... or both?