Towards Benchmarking Foundation Models for Tabular Data With Text