16 bits is not far more than we'd ever be able to use. It is about enough if things are just right but when we add in a few real world concerns involving processing, level and the like 24 bits starts to look preferable in order to remove any concerns. 32 bit floating point would be perhaps better again these days since this gives almost 24 bit precision while removing concerns about level. The hardware limitations that lead to 16/44 becoming a standard disappeared decades ago.Yes I agree, the noise floor is lower meaning greater resolution of even lower level signals - that’s what dynamic range is. However, if we’re playing music at normal levels it’s not possible to hear any more of this resolution, as 16 bit audio already has far more dynamic range than we’d ever be able to use for music listening.
44 kHz sampling does ask a fair bit for the filtering to be clean upto 20kHz as mentioned earlier. 48 kHz would ease this while 96 kHz would allow concerns about multiple stages of processing to be dropped. 32f/48 playback (if it is expressed like that) or 32f/96 for playback and recording/processing looks about right to me.
The problem with high sampling rates is that the signal may contain energy at inaudible frequencies above 20 kHz which could then interact with nonlinearities in the playback chain and be aliased into the audible frequency range and heard. I suspect if these inappropriate sampling rates do become common the more informed folk are likely to start using 'scratch' filters again!
Out of curiosity do turntable enthusiasts these days tend to use 'rumble' filters or have they disappeared along with tone controls? There can be similar issues with nonlinearities aliasing infrasonic signals into the audible range.