I've read that ZFS is less safe than other Linux filesystems if you don't use ECC RAM, because it assumes that there are no memory errors and therefore doesn't provide a tool to repair a filesystem corrupted by such errors. Is this true?
It's not true. That's basically ancient forum myth, alongside the also incorrect "ZFS needs 1GB memory per TB of HDD" nonsense that has thankfully mostly died out finally. ZFS makes no additional assumptions when using ECC vs non-ECC memory.
It is theoretically possibly to construct a scenario where evil ram does all the exactly right things needed fool ZFS and corrupt your filesystem. Any pearl clutching about this thing which has never happened somehow also ignores that every filesystem is going to get corrupted.
In reality, while ECC memory is always nice to have, it's no more required than any other filesystem. Though personally now that amounts of +32gb are common, I generally prefer error correction/detection over ultimate speed these days. Though ironically ECC memory is actually really nice to overclock, because I can actually just check my logs and prove if my system is actually stable.
There so many actual dangers to your data in comparison that it's laughable. The biggest one being you. Followed by hardware failure, malware, and genuine ZFS bugs. I'd stay far away from raw sends of encrypted datasets in ZFS for a while, there are edge cases that haven't been resolved yet.
