aside from spending a few extra bucks and potentially looking like audiophool, are there other downsides to going with 32-bit?
Well, aside from probably not being useful, one of the downsides, if you take a look at datasheets, is that affordable 32-bit DACs (eg: PCM1795) have actually worse specs than the good 24-bit ones (eg: PCM1794A) for similar prices. And as I mentioned, better 32-bit DACs are pretty expensive.
Another point related to what Bud asked you: what will be the typical sources? Most common input sources do not support 32-bit anyway, so you'd be using a 32-bit DAC with only 24-bit (at best) samples, with again usually worse specs. What would be the gain?

As to be future-proof, I dunno. No one can predict the future, but I kind of doubt 32-bit is going to be standard any time soon. One reason is that we're getting very close to what's reasonable on a physical level here when it comes to audio. You'll probably have ample time to make a rev. 2 of your device if you ever see signs that it's going to happen.
One reason to use a 32-bit DAC would be to avoid having to reduce quantization if your input source is natively 32-bit - because no matter how good the reduction is done (with proper dithering), it's never going to be as good as the original. But as I said, I don't know of any source currently that is *natively* 32-bit, at least from the standard ones.
Anyway as I said, if you do things really really right, you may be able to get something like 110/120dB SNR with a *good* 24-bit DAC, but only up to its output stage. The power amp stage - that's another story, and something you should definitely work on to get any benefit from using a DAC with better than 100dB SNR (or even 90dB...!)
Can you recommend any good resources for learning more about layout practices for good SNR?
Hmm not really, but I'm sure some other people here can point you to some books or something.
I've been learning about EMC lately and seeing conflicting advice from EMC people saying never separate ground planes even when datasheets recommend separating analog and digital.
Ah yeah. What I can tell you is that most often the better way is to simply use a single solid ground plane and NOT split ground planes. The latter should be used sparingly, only for very specific cases, rarely when there is high speed digital stuff involved (which is the case with DACs). Just don't do it here. But you should definitely route your signals so there is minimal crosstalk between the analog and digital ones.