Fix loading IntervenableModel for its subclasses

[PinetreePantry]

Fix issue #45
One way is to replicate Pyvene's load() function in Pyreft. That is not great, because later if we modify Pyvene's load(), Pyreft()'s load() could be left behind.
Other way is to replicate all parameters of Pyvene's intervenable_model, which is what this PR does.

[frankaging]

Hey @PinetreePantry can you check memory overhead? please make sure this is only assigning pointers and creating another object (doubles the memory)

[PinetreePantry]

I think we are okay. We are not performing any large copies here, just pointing all the variables in the new class to the old class. The IntervenableModel object is being thrown away anyway after the load.