When the audio source provides 24 bit samples, don't bother to convert
(lossily) them to 16 bit before jack's floating point conversion - go
directly from 24 bit to float.