It is particularly odd in the case of open-source clients (or indeed any client that runs outside of some very locked down hardware) because a) there's nothing that prevents the user exfiltrating keys anyway, and b) attestation also means relatively little for such an implementation.
Yes, the problems are obvious and the spec authors definitely know & understand the issues. Their refusal to have a public discussion about it indicates they just don't care, and their maintenance of a "naughty client list" shows Passkeys are intentionally hostile to user freedom.