Data integrity enforcement in Clean architecture: Domain or Data Layer?

a person needs to have at least a pet. Because of this rule, we can't have a person without any pets.

This sounds like a invariant which belongs to the Person domain concept.

So to keep the a Person data consistent, I would expect it to have a function where you can remove a pet if it belong to the person. Then in that same function you can check if there are any pets left and if not, delete the person also.

Like in this pseudo code:

public class Person 
{ 
   public Collection<Pet> Pets { get; private set; }
   public bool Deleted { get; private set; }

    public Result RemovePet(PetId petId) 
    { 
        var pet = Pets.Find(petId);
        if(pet == null) 
        {
            return Result.Failed(PersonErrors.PetNotFound);
        }
        
        Pets.Remove(pet);
        
        if(Pets.Count == 0) 
        {
           Deleted = true;
        }
        
        return Result.Success();
    }  
}

Now you persist this Person domain model via the Person Repository and the repository should figure out to which table the data should be persisted.