I interpret that "currently" there would be some methods for nodes to exclude these "fake public keys" from their UTXO set already, but if the validation was changed, that would not longer be possible.
Kind of-- half of fake pubkeys are indistinguishable from ordinary pubkeys to begin with. But the other half aren't and so outputs that are actually unspendable due to a detectably fake pubkey could be excluded. If there were some relay rule running this check then none would be detectable anymore.
When an output is unspendable no one can tell if you've pruned it or not, so you're free to use whatever complicated schemes you want to tidy your own utxo set
so long as they have no false positives for unspendability.
Of course spammers, if they really want to be sure that their data remains on the blockchain,
Well that isn't what they actually want. They want to hype their worthless token over some other worthless token, once they've pocketed the victims money they sail off into the sunset and the long term is irrelevant. So it's just a marketing point.
could already now develop a completely "safe" protocol,
Nah, because anyone paying attention would know that the existence of utxotree shows that no such safe method can exist, even using spendable outputs.
Bitcoin isn't a data storage system so surprise surprise you cannot use it to safely store data long term.The other method we have is to enforce transaction (not block!) pruning as an optional command-line option, by using a bloom filter to completely ignore Bitcoin eater addresses, invalid keys, OP_RETURN.
Bloom filters have false positives, you can't do that because someone will come along and spend one of those false positives and then you reject the valid block and go off onto a fork. (Also the victim of the false positive would be pretty cross their coins were destroyed if that fork were ultimately popular!). OP_RETURNS are all safely excluded from the UTXO set today because they can never be spent. There are other outputs that are undependable which could be detected but it hasn't historically been worth it.