not verifiable (nutrition science is not settled, to say it mildly)
Yes, it is relatively rare and can be fixed manually but doing it manually brings no significant benefits over doing other QA. And at least I have enough QA ideas to last for several lifetimes. With User:Mateusz Konieczny/bad vertex - OpenStreetMap Wiki being latest published report.
yes, but global community is far less bot-happy so I am proposing first for USA and Poland, and later I will propose it globally
this way I can point to a success story if everything went 100% fine, and nearly everything (code, documentation) can be reused at minimal extra cost anyway
and sometimes global bot edits like this get rejected (despite making sense) and getting accepted in more local regions
also, I feel I run out of quota for global forum posting for now, and I will wait a bit (and have some more important ones to run first)
AFAIK we have ATYL in OSM, so you would need good reasons to remove someone else work from OSM. I see this different eg. form removing some import-meta data, where no one put actually effort in.
Overall I think it’s a helpful thing to know whether the restaurant eg. serves just processed food or “natural” food. I agree with you, that the diet:healthy is maybe not the best key. But instead of deleting it, I would think it’s better to reach out to the mappers, who used it and find a better, less subjective key.