But it's not by anyone at all. It's only by users that have unlocked their database. I really don't see the attack vector here.
It's not like the Apple Keychain at all, since your interaction with Keychain is very different than KeePassXC, which makes the locked vs unlocked state very explicit (and you're almost always auto locking anyway), whereas Keychain is something happening in the background that sometimes prompts me for my password/fingerprint. I have no idea what the state is there and would be very annoyed if someone could leak all my secrets just by accessing my computer.
With KeePassXC I'm always aware if it's open or not, because I can't use it without knowing that, and I had to make a very explicit opening of it. Because it uses local files and not the cloud, it's very important to me to be able to import and export the contents. Without that ability, I will lose access to my passwords.
It doesn't mean it should be easy to do, but it's also completely unacceptable to make a requirement like "users are forbidden to access their private keys".
> by anyone at all
What do you mean by anyone at all? By the owner of the private key. Not by anyone.
If I log into my computer and turn my private key into a plaintext blob, as a file or a Python object or something on a USB stick or a QR code that I photograph, then anyone who happens to have compromised my computer at the time has my public key, too. Even if I subsequently fix the compromise, they still have my public key.
I do not want this to happen.
If someone has sufficient access to your computer (like being able to keylog and stuff) - it's somewhat late to worry about keys being compromised.
If a passkey can have its private key exported, by anyone at all, then this property is lost. I do not want access even to my own passkey private keys!
I certainly don’t love having a few gatekeepers in charge, but the protocol (currently?) does not really support a good alternative. And doing better is hard!