Exploring Phoneme-Level Speech Representations for End-to-End Speech Translation