I measure them all day long without need for special equipment.
1) "the outputs are floating" *really means that *the load* is floating; or to be more precise, no load terminal is grounded *but* each of them receives a signal which *still is* referenced to ground.
It isn't as if signal came from a floating transformer winding referenced to nothing !!!!!!!!
So I just clip my scope fron one hot terminal to ground (nothing special here) and drive it to barely clipping, then back up a little.
Then check the other terminal, same thing.
If I have a regular run of the mill dual trace scope, can check both sides at the same time, go figure, always referenced to ground and with no wild grounding or floating setups needes, at all.
Then I measure both voltages (referred to ground) and add them up. Piece of cake.
No unsafe ungrounding, no diff probes, nothing special.
2) suppose you are measuring audio power with, say, a 1KHz tone.
suppose your amp switching frequency is, say, 150KHz.
Use an RC filter at, say, 15KHz , such as, for example , 1Kohm/.01uF .
It will be easily driven by your power amp, will not be loaded by your scope's high input impedance (usually 1M) , will attenuate the 150KHz signal by around 20dB (or 10X) which should be more than enough.
3) why unfiltered Class D? Not a good practice hitting your speakers with that kind of HF energy, not to mention FCC will kill you for polluting the airwaves.