IANAL…
I’d say it depends on the case. If a user mentions personally identifiable information in a post, or if the conversation flow mentions the user repeatedly, making it obvious who is talking, the anonymization function won’t work. OTOH I think the terms of service should make clear that if a user decides to leave, their public contributions bear a non-exclusive license (e.g., CC-0 or CC-BY-SA) that will protect the site in case of a ‘divorce’. That said, CC-BY-SA might be even trickier, since the BY clause means you want to keep your name on it, and of course, if you decide to leave, you only lose access to those words you already shared.
I think it’s important to realize that when an individual participates in a conversation, they become part of a trans-individual activity that’s larger than any single participant, where each gets nourished and inspired from the whole, and engage accordingly. Given that, it’s difficult to be willing to remove that from the group, since it would create a hole in the conversation that, to me, is sabotage. If people cannot deal with their words, anonymization is probably the best solution, but it also has drawbacks: if the person actually reverts his decision, the de-anonymization is nearly impossible or at least very costly (you would have to go through all anonymized posts before the decision was made and grant them back to a new account.)
It seems to me that, as often with law, a sane balance cannot be found within the conditions of the law itself, but requires attention to solve the issue case-by-case. If you consider the upcoming law to limit subsidies to larger farms in order to support family farming against industrial farming, then the Czech will cry because they have so much larger farm lands than, say, the French: it’s not because they have larger industrial farming, but because the socialist economy imposed large farmers cooperatives… So one size fits all is not how things work in reality.
One thing we should take care of in priority for European instances of Discourse, is to transpose the default ToS and Privacy Policy that ship with the product to European law – GDPR-compliance is a good pretext for that, but certainly not the only one. I’d like to work with ToS;DR on this to provide a default Class A ToS for Discourse instances in the EU. I started some time ago to think about it, but it requires IMO – besides some legal knowledge – a participative process and so far not many people seem to be interested in solving this issue. I’m glad this discussion exists here!