If you are doing automatic error correction then why not just pregenerate the entire thing...? They may have errors there on purpose, or as mastering error.
You can't pregenerate the entire thing. P-channel layout, EAN sectors layout, philips/sony gap types, etc. - these differ if you simply generate the subs from cues. And a cleaned (only from 1-bit errors!) sub is way better than a 1-pass "dirty" one, because it is verifiable. Mastering 2-bit (and 2+ bit) errors are always untouched and preserved in this case (for example, on the PCECD F1 Circus Special - Pole to Win (Japan) [HRNB20505-4FAAT] disc or on the SS Rockman X3 (Japan) [T-1210GP-00798 2M1 C 64] disc).
Mastering 1-bit errors are common, yes. As for intentional 1-bit errors, I only know the PSX discs with 0x01, 0x02, 0x03 patterns in R-W channels and yes, in this case you shouldn't apply 1-bit correction to R-W channels, because it wipes all the 0x01 bytes. But, hey, dumping process should never be "blindly" automatic, but always under a human guidance.
Cross-comparing multiple dumps sounds good in theory, but you need a lot of copies of a game for that, might not be as simple for expensive games or one of a kind prototype discs.
Well, AFAIK, Rawdump admin tried to examine that and he was able to quickly sort out all the matching errors from 2 different discs dumped on different drives, though, it's always hard to judge, because you never have a "perfect" dump with all the mastering errors there to compare with.
Again, noone says you shouldn't keep "dirty" subs for the documentation purposes. But for the general public release cleaned verifiable/verified ones are better. And instead of storing the subs externally, like you've proposed above, I'd rather think about the way to keep the list of cleaned errors externally instead (maybe in some kind of xml file, maybe in the log file, maybe in the chd metadata).