Feed-Forward Bullet-Time Reconstruction of Dynamic Scenes from Monocular Videos