I may be wrong, but I'm fairly sure that 32bit floating-point processing is not going to pull the same info from analog tape master made at 16bit as it can from a 24bit fixed point master of the same tape.
32bit floating-point is not 'pulling more information out' of the analog tape, its simply providing increased processing precision for calculations made after the information has been digitized. That processing precision ‘headroom’ is only needed while doing the calculations. It’s unnecessary to either capture that range or output it after the calculations are made, it simply avoids ‘rounding errors’ during the processing.
Expression of dynamic range for a cassette, or any analog tape for that matter, is different than the same measurement in PCM. Both are based on the point the noise floor is reached, however with digital it goes from signal to complete noise at a fixed point, and clips to all distortion at a fixed point. With analog, the same cut offs apply for SNR and dynamic range, but there is musical information above the clip point and into the noise floor on analog tape not included in the numbers, but audible.
As long as the digital dynamic range fully encompasses the entire dynamic range of the analog source, including those non-linear distortions at the extremes of the analog range, it can fully encode all the information. The digital range just needs to be as large as the
entire analog range, with levels adjusted so that full analog range fits comfortably within the digital range. Just use a digital range that is large enough to include any ‘soft-clipping’ at the top
and decay into noise floor at the bottom. Inclusion of that extra bit of range still probably fits into 16 bits.
Dither is a good way to think about this, as it allows sounds to be audible and decay gracefully as they fall below the noise-floor, without a sharp digital cutoff point. It’s simply the addition of analog noise. To retain the same measured noise floor when introducing dither, the digital range needs to be extended by an additional bit to encompass the dither. The digital range then fully encompasses the analog range including the dither and the analog sounds which can be heard to decay into the dither below the analog noise floor.
The digital dynamic range only need be slightly larger than the analog range to completely capture all the information; it doesn't retain any additional useful information by being considerably larger.