1. I see. It looks there's major trade-off between preserving backward compatibility (not many things could be added) or breaking backward compatibility (less adaption/usage).
I don't think so. OP's proposal as it stands is 100% backwards compatible. If I generate a new 15 word seed phrase with the 32 bit versioning system, old software will still see the entire seed phrase as a valid BIP39 seed phrase and will recover the exact same wallets (provided of course I know my script type/derivation path, since old wallets won't be able to interpret the 32 extra bits). Any legacy seed phrases will still be recoverable by new software, provided there is a simple check box to indicate it is a legacy seed phrase so the software does not try to interpret the first 32 bits of entropy as versioning data.
Yes, i'm aware OP's proposal is backward compatible. What i actually wanted to say are,
1. With backward compatibility, that means there's not much room for more improvement/change. In this case, some people may think "Why bother accepting/using this proposal?".
2. By breaking backward compatibility, it could hamper adaption or usage.
Perhaps you could even stipulate that the new seed phrases must be either 15, 21, or 27 words long, and so any seed phrase which is 12, 18, or 24 words is immediately known to be a legacy seed phrase.
As reminder, BIP 39 allows entropy from 128 - 256 bits with multiple of 32 bits. It means you can generate 12, 15, 18, 21 or 24 words of BIP 39 mnemonic.