In my previous non-Bevy OpenGL projects, I always extensively use texture atlases for almost everything to increase performance (since the glDraw* methods are relatively really slow). But when I look into Bevy codes, most of the codes and external crates directly use Image, and even when they use TextureAtlas-es, it's really minimal (benimator for example, uses a single TextureAtlas for individual animations).
What's the reason behind this? Won't it be much more performant if more sprite renderings are batched? And does the performance cost really matter in Bevy (or wgpu, I guess)?