LAMASSU: Streaming Language-Agnostic Multilingual Speech Recognition and Translation Using Neural Transducers