ActivityImplPtr are all over the place, and the manual refcounting is
dead. Most tests still fail, for some reason that I don't understand
yet. That's a WIP...
Notable point:
MC wants to retrieve the pointers to the ActivityImpl in a remote
process, even if their type is ActivityImplPtr (that is,
intrusive_ptr<ActivityImpl>).
The good thing is that in this case, the raw pointer is stored in the
simcall parameters, and the *Ptr type is reconstructed by the marshal
function.
So I just have to generate the <simcall>__getraw__<param/res>()
functions that return the raw pointer without building a *Ptr type. It
compiles, but the MC is broken ATM so maybe it's not sufficient.
I'm committing anyway because I want to add simpler tests for the
communications that would make the debug of this feature easier to
achieve.
