Inlining of variables binded to unit

Constructors without arguments are now inlined in extracted code

This commit fixes some bugs for coercions.
1. Applying coercion should work correctly w.r.t. unification
(see  modules TrickyPolymorphic(Alpha|Beta) in bench/typing/good/coercions.mlw).
2. Union of two coercion maps now works correctly w.r.t. adding the same coercion twice
even if it is transitive closure
(see module SameTransitivityCheck in bench/typing/good/coercions.mlw).
3. Coercion error printing is now listing all components of conflicting coercion
(see bench/typing/bad/coercion_cycle3.mlw for example).

In this version, we still just lookup at head tysymbols to decide whether a coercion
can be applied. This can be improved by taking into account also the types of arguments
to decide earlier that a coercion cannot be applied.
For instance, given

type t 'a
function f (t int) : int
meta coercion function f
goal G: forall x: t bool. x = 42

results now in an error message
This term has type t bool, but is expected to have type t int

We can do better, detecting that a coercion f cannot be applied to [x] at all.
To be done.

Forgetting local variables identifiers

a slightly different representation of coercions
(using trees instead of lists)
bench files for coercions

Now adding the very same coercion is accepted (useful for multiple/transitive use imports)

Minor

Registered exceptions in file coercion.ml

IDE option to show coercions in formulas.
By default this option is set to true

Composition of coercion functions is now labelized

Started a module to contain operations specific to coercions

Treatment of top-level recursive definitions

Erasure of ghost code in statements sequence

div -> ( / )

Minor changes in compile.ml and ocaml_printer.ml

Adding support in drivers for Zarith and the ocaml.mlw file