Let's start where we agree. The headroom of 16bit is more than enough. We're good there. Again I said work flow. With 16bit you have to record, for safety, expediency and cost at levels typically between minus 20 and 16 dbfs. Smack down the middle we are only using about 13bits. So yes you would normalize to 0 dbfs thereby using the full dynamic range
but this has its own set of potential problems nor would it in either 16bit or 24bit still obviate the requirement of some judicious compression on a case to case basis. These problems like the potential of adding artifacts from too much processing can be avoided by recording to 24bit at levels where you are equivalent to the usable dynamic range of 16bit. No need for too much added processing. I think we are still in agreement at this point. What would be crazy is to try and utilize the full dynamic range of 24bit for obvious reasons, no system can probably handle it but more importantly we humans wouldn't and couldn't. Here I am SURE we are in full agreement.
My recording background isn't in music, it's in cinema and TV, but the tools are the same. The final outcomes are also the same in that the material is prepared for an intended use. So I am not on the anti compression or anti normalization band wagon as they are useful tools if the goal is delivery of the material to the most common denominators. In cinema attention is paid to the quiet rather than the peaks compared to music as we add tons of low level elements that viewers pay no attention to like foley elements. Take out the rustling of the clothes, the footsteps, the wind or even the room tone and the experience becomes weird and unnatural. To a certain degree this is true of music too.
Wait, where don't we agree again?
I think I am now seriously off topic since this is a DSD thread and I am a DSDunce. I only ever recorded with PCM and the highest I have used is 24/88.1. The greater majority of the time was at 24/44.1.