It won’t be “bit perfect”, but there will be no audible loss.
24-bit integer has a dynamic range in excess of what even an expensive professional sound card can produce. 32-bit float has even better dynamic range, so you are only “losing” audio information that is too small for your sound card (any sound card) to be aware of.
The algorithm, I proposed, turned being a total failure, as shown on the attached wave plot.
I calculated on excel the mono samples from the first stereo 100 samples and plotted the results; M1 by using the conventional addition and M2 by using the one I proposed.
