testnet actually has the same powLimit as mainnet, so this would change results only on regtest, but doesn't it make more sense for difficulty to always reference the same minimum value and regtest just allows it to go less than 1?

Hm, I didn't realize that (testnet had same powLimit as mainnet). I do know that when running on an experimental network with a different pow limit we got very weird values which go against the comment about being a factor of minimum difficulty. I guess it comes down to, "do we care about that case?"

NACK, let's not try to redefine difficulty for each network. It's okay if some networks might have <1 difficulties.

I believe that in the first testnet iteration, at least in some versions of the software, the powLimit was defined as 2x higher than the mainnet one, and 1 testnet difficulty corresponded to 0.5 mainnet difficulty in hashrate. It was pretty confusing, and I think it's indeed easier to just everywhere treat 1 difficulty as 4295032833 hashes.