Diffusion Meets DAgger: Supercharging Eye-in-hand Imitation Learning