I expect 32 bit to be useful at low levels, not really for high level gain setting. 32 bit does unchain you from any gain change constraints in post, you can crank up -50dBFS signals with little/no penalty.
If you’ve got some imaginary input with supersonic specs. Just because you’re recording at-50 dB doesn’t mean the noise floor is not still around 115. No advantage over 24 bit
You are right the analog noise floor remains the same regardless of 24 or 32, and it may be poor/average in this product. If they are deriving a pretty clean 32 bit word (remains to be seen), the bottom bits are so far down that you can set gain after the fact, which may be better than what the onboard preamp contributes if cranked up. Digital gain on a low level signal at 32 bit float with a 32 bit word may contribute less noise than a Zoom preamp cranked way up. It may be the same, but more free of dither noise artifacts for dialog production in film/TV. It's apples to oranges in lots of ways, but it may give a bit of commercial advantage over the production costs of quieter analog input stages while easing the requirement to worry so much about input levels.
Typical ambience in a quiet room with most microphones is generally enough to mask most analog noise floor anyway, unless that noise floor is significant. Most new recorders have pretty respectable noise floors.
My MOTU 16A converter (24 bit with 32 bit output bus, +24dBu max input = 0dBFS) shows a noise floor no worse than -137dBFS (-113dBu) at 16kHz in an analog wired loop test from DA to AD, -143dBFS (-119dBu) at 4kHz where the ear is generally most sensitive. It appears to be defined by the dither curve, when you look at various sampling rates up to 192kHz. That compared to the Zoom F8n at least is a much bigger difference, and easily trumps 24 versus 32 for practical purposes, and that basically agrees with your point.
I had an F8n (interface mode) and the 16A connected as an aggregate device into a DAW, and used a Sytek MPX-4Aii driving the 16A, set the Sytek for max gain and matched the F8n gain to it. Then looking at the noise floors, the F8n was 2.5dB to 1.2dB noisier from 1kHz to 16kHz, but 4.5dB at 250Hz, 7.5dB at 100Hz, and 10.5dB at 30Hz.
What I haven't done yet is measure noise comparisons at minimum gains, will have to do that.
Anyway, spitballing here, lots of contradictory 'what ifs' and unknowns. I definitely don't feel any need to upgrade, but it may be something to look forward to when the time comes.