Presentation context always has GL_DRAW_FRAMEBUFFER_BINDING as zero.
There is no need to bind the default framebuffer constantly.

According to Nsight this was using ~0.7ms per frame and it broke
renderdoc captures.