Hmm, I did a lot of recording during the 16-bit days, and I beg to differ.

A 16-bit linear signal gives you 96dB of dynamic range. That sounds like a lot but unfortunately real musicians get excited and do things like yell into the mic or hit the strings very loud - things you can fix in post-production, if the track isn't clipped.

In the earliest days, there was no over protection, so if an over occurred, the track was simply trash at that point. Quite soon, soft clipping prevented the worst of it. But as an engineer, you still had to be very careful to turn the mic down enough that real "in the red" overs were impossible, while having it up enough that the noise floor was invisible. And remember that those tracks also get EQ'ed and processed, which in some cases makes that noise floor jump right up.

When 24-bit recording came in, all of that went away. Even though the last few bits on 24-bit recording are generally noise, it still meant you could just put 30dB or 3 bits of headroom above what you believed was your loudest noise, without having any detectable noise floor.

As for float vs integer, well, as someone who has been writing DSP code for over 40 years (mainly for fun), I'm always relieved when I'm working in float because arithmetic overflow is impossible. Even a tiny number of arithmetic overflows will be quite noticeable in the output track, as ticks or clicks.