RAD-TTS: Parallel Flow-Based TTS with Robust Alignment Learning and Diverse Synthesis