A Non-autoregressive Generation Framework for End-to-End Simultaneous Speech-to-Any Translation