Languages in Multilingual Speech Foundation Models Align Both Phonetically and Semantically