I don't see why it cannot be increased to the extreme, let's say to 2100000000 bytes (2.1 GB).
It could be. But: non-upgraded nodes should not process all of that data. And, you don't have to increase maximum block size specifically. Instead, quantum commitment size can be restricted.
For example: if quantum signatures will take 50 kB per signature, then let's make it as a commitment limit per sigop. Then, for a block sigops limit of 80k, we would have 4 GB limit per block. And if there would be any need to do a downgrade, then coins could be moved in a way, which would be understood by old nodes.
In this case, each and every OP_CHECKSIG call will check two things: one, which is ECDSA correctness, and another one, which is quantum proof correctness. Then, private keys can stay as they are, and R-value of a given ECDSA signature can contain SHA-256 commitment to any quantum signature, which would be handled only by quantum enthusiasts, while everyone else could enjoy 4 MB limit, and see just regular ECDSA signatures.
If hashrate majority will run quantum-resistant version, then the block size limit would be just 80k sigops * max quantum signature size. And then, for different quantum proposals, that limit can be different, while keeping the same sigops limit per block, so handling roughly the same number of individual on-chain users per block.