All audio processing inside Cubase is done at 32-bit floating-point format, no matter which format is selected for export nor which format you decided to set your project to record audio. It’s just final step of the export/bounce, which truncates this 32-bit stream into whatever bit-depth you have chosen.
Wrong. Cubase truncates samples to whatever bit-depth you have chosen. No dithering needed. Do you even know, what is dithering?