Add type inference for argument and return types

[Matt Giuca]

A lot of recent work has gone into inferring variable types so variables no longer require declaration (bug #483082). We can now go a step further and not require explicit argument and return type annotations (Mars will be a fully type-inferred language).

Due to the way we have set up the type checker, this should actually be very easy -- it should only require changes to the parser. Any argument/return which leaves off a type should be bound to an *unnamed* type variable. The fact that it is unnamed means it will not be made rigid, and thus may become bound. That should be sufficient to get type inference working.

[Matt Giuca (mgiuca)]

In parsem, algorithmically generate a unique name for each unnamed type variable, from the set {"a", "b", "c", ...}. Otherwise, you will start seeing functions with types like t_1 -> t_2.

It is important to do this *after* the set of rigid variables is generated (it is based on which variables are named, and we don't want these variables to become rigid just because they have a name).

[Matt Giuca (mgiuca)]

Or just fix the type printer to generate nice names for unnamed type variables (a, b, c, ... instead of t_1, t_2, t_3, ...), making sure that no generated name collides with a named variable.