The Thesycon and XMOS USB drivers support both 32-bit float and integer. I don't know if there's a distinction in the Windows sound settings.I have seen 32bit Integer output devices, but not FLOAT. Been looking for some just out of curiosity, but no luck.
RME devices accept 32-bit float natively on macOS. You can see > 0 dBFS overloads on the device, and can attenuate on the device itself to avoid clipping at the analog output without damaging the signal.
WASAPI operates in 32-bit float, so the mixing should not clip. In another thread, a test was done with +3 dB in Foobar and -4 dB in post-mix. There was no clipping, indicating that CAudioLimiter is not applied until after post-mix.Fix has to be applied in pre-mix because the distortion appear to happen during mixing process. That is also why it does not matter what parameters the output device has.
However, Foobar was outputting 32-bit float. I don't know what happens if iTunes outputs 24-bit integer.