Speech foundation models on intelligibility prediction for hearing-impaired listeners