Would this be possible?
It is better to make a regular soft-fork in that case:
https://groups.google.com/g/bitcoindev/c/p8xz08YTvkwI can imagine the "simulation algorithm" to be enormous, like the gigabyte-big ZK provers.
It shouldn't matter, because you shouldn't put everything on-chain. See:
https://bitcointalksearch.org/topic/really-really-ultimate-blockchain-compression-coinwitness-277389 (this topic contains the word "witness", but it is not about Segwit).
Or are there unsurmountable challenges?
Even if ECDSA will be broken, then still, there are scripts you can do, which could be safely used, even in that kind of scenarios. For example: "OP_SHA256 OP_CHECKSIG" is one of those Scripts, where you have to put a message, which will hash perfectly into x-value of the public key, and will pass Schnorr signature verification.
Another example is "Pay to Proof of Work", when you require a DER signature below N bytes.
So, even if OP_CHECKSIG will lose its original meaning, then still, it will then be just a calculator, working on 256-bit numbers. But: it will be possible to mount another challenge, where you would need many OP_CHECKSIGs, to move the coins. And they can be wired in a way, where knowing the private key will give you no advantage, because the challenge will require solving dependencies between keys, and not the keys alone.
Was this perhaps even discussed in some technical forum or mailing list already?
Well, there are some posts. Here is another one:
https://groups.google.com/g/bitcoindev/c/SPmrzARLMFU