“Some person with physical access to the computer” absolutely is part of the threat model.
Some person, yes. The owner, no.
The standard way you distinguish "adversary with physical possession" from "owner" is "do they have the owner's passphrase".
I am not familiar with Macs so I might have been wrong from the very start, but if the owner put the key in the secure enclave then the same user should have no problem at putting the same key in the secure enclave of another Mac. If the user can't write into the secure enclave, then the Mac is protecting the data against the very same user that is supposed to own those data. It is protecting against wrench attacks too but what's the most frequent scenario?
I mean the most frequent scenario is that the person trying to unlock the device is the one who owns it, yet we still have account passwords. You have to protect against the worst case. There is not a trusted enclave in existence (Apple, Intel, etc.) that will let you read its contents. That’s the point!
The actual key cannot be exfiltrated from the Secure Enclave. All the encryption/decryption is done inside the Secure Enclave. Knowing the owner’s password is just enough to let you pass data through the Secure Enclave to do the appropriate encryption/decryption operations.
So, you can’t copy that key over to a different Secure Enclave. If you could, then it would no longer be a Secure Enclave.