This is needed in order to allow the tuple caster to accept any sequence
while keeping the argument loader fast. There is also very little overlap
between the two classes which makes the separation clean. It’s also good
practice not to have completely new functionality in a specialization.