Add Constant Propagation pass after MIR passes

[Leo-Besancon]

As raised in the comments of #424, the MIR pipeline shows a regression of Constant Propagation in some cases (in particular, for Exp operations where powers are indices in a list comprehension).

We should either re-implement a Constant Propagation pass on the MIR, or, if more suitable, better handle the Exp operations (as the scope of the issue is limited to specific cases).