Design: have data implementation for each type

[renaudpawlak]

1. every data should be a DataReference
2. depending on the type/context, we will switch the implementation (ListData, RawData, ArrayData, ObjectData)

Note: this opens potential optimization.

• keep all the instances until a data modification invalidates them
• if no data modification, we can switch back and forth for the right implementation to maximize peformance

=> note also that DataReference are flowed in the design... for instance:

l = [["a"],["b"]]
s = l[0]
swap(l[0], l[1])

This changes the value of s when swapping because s is a reference (i.e. when s is not a primitive object). IMO, the reference system has to be entirely rethought.