Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps