Merge: Optimizations in nitvm

This PR introduces some change in the implementation of object mechanisms in the nitvm.

Now, the dispatch between several mechanisms is made directly in the AST nodes instead of in runtime mechanisms themselves.
This PR is preliminary work before the implementation of complex optimization systems, but now the nitvm is sligtly more efficent than before because the choice between implementations is only made once (the first time).

I use a workaround in AST nodes to avoid conflict with refinements between the interpreter and the vm.
In the future this will need to be well implemented, maybe by creating a sort of ```CallSite``` for subtyping test and attribute access and store informations into them instead of into the nodes.

Pull-Request: #1120
Reviewed-by: Alexis Laferrière <alexis.laf@xymus.net>
Reviewed-by: Jean Privat <jean@pryen.org>
Reviewed-by: Alexandre Terrasa <alexandre@moz-code.org>