These and some other comments make me think you don't have proper perspective on resolution.
First, "at least some DACs use 50-bit words internally": For DSP processes, we often need to retain precision, temporarily, that is far more that the actual digital to analog conversion. This is especially true of filters. I'll make up a dumb process here as an example, but it's not far off from what goes on in practical filters. n / 10... x 10. We should expect to get n back. But if n is a value 0-99, feeding a process that accepts 0-99, if our divide and multiple doesn't retain fractional values, then an input of 24 would yield a value at the output of 20, a huge error. If that math accommodates a range of 0.000-999.999, for instance, math like this comes out fine.
But for a gain change, we don't need this level of precision, because there is no penalty for losing unused bits. If we have 24-bit audio (fixed point, roughly representing -1 to 1). Now, the gain value could be 24-bit, because we're already working with that resolution, or it could be 16-bit or even 8-bit. Whatever it is, the results of the multiple yields (basically) the resolution of the digits added together (9 x 9 = 81, for instance, it take 2 digits to multiple two 1-digit values). If you don't, you lose the least significant digits. But, we're putting the output into a 24-bit DAC anyway, so we're going to toss those bits. We could retain one for dither, for folks who must dither 24-bit, but the point is we doesn't need Pauls 50-bit words just for gain. For filters, especially recursive, sure, but not just the gain.
Second, "With 32-bits, I would have 16-bits at -96dB of DAC volume": Again, recognize that 24-bit DACs are the limit of practicality, limited by physics (and fortunately, our ears anyway). And do you understand the implications of having 96 dB of volume control range? Basically, it means a range of being loud enough to cause hearing loss over time to being quieter than your own breathing at rest. I'm not saying it's bad to have that range, just giving a rough reference to what it means. More importantly, you seem to want this control down to the quietest sounds in the quietest rooms, while retaining the details of the recording that are another 90+ below that barely hearable level.
Two problems: The voltage levels represented by the bottom ~60 dB of that are below the floor voltage noise level of electronics. (No, you can't make it quieter.) And your ears are already less sensitive than that level anyway. For sound levels, if you're listening at a very loud 96 dB SPL, so that 0 dB SPL is the floor, and you turn down 96 dB so that that floor is now at -96 dB SPL...well, in a typically "quiet" room, you won't hear 0 dB SPL, so you can imagine what a stretch it is to think you might be able to hear another 96 dB down from that.
My point again, is that 24-bit is not only better than the best we can do (the error in the bottom bits is relatively huge, constrained by physics), it also outpaces our ears by a wide margin. And the only way you can make the latter not true is with horrible gain staging—an enemy of both digital and analog audio path, equally.
As an aside, I never see audiophile talk about this, but you don't want the loudest amp and speaker combination that you can buy. Say, for instance, you buy amp and speakers that can attain 150 dB SPL at you listening position. Well, since you don't want to be in pain and then hear very little the rest of your life, you might dial your preamp back, say 65 dB. But, the preamp still puts out its minimum noise floor, turning it to zero will not stop that. (The amp too has a noise floor ahead of its massive gain, so it's not a fault of the preamp, it's a fundamental limit of electronics, and it doesn't matter which stage you blame it on.) Ideally, you want the amplification to be louder than you will listen to at the loudest (you want some headroom) when your source is turned all the way up. Then use digital (or analog) attenuation for non-max levels. That will give you your full dynamic range and lowest noise floor.
Lastly, if you're worried about gear that doesn't use enough bits for the gain multiplication, you're fundamentally worrying that someone out there might make crappy gear. Yes, people make crappy gear, analog and digital. For the most part, the worst offenders go out of business, and that encourage the rest to try to not make crappy gear. And that includes cheap analog level controls that get noisy and scratchy, something that won't happen with digital volume control. Except for in crappy gear.