Allow the type checker to accept empty values

[gzanitti]

No description provided.

[chriseth]

Oh I wouldn't introduce a new type, this should just be the empty tuple.

[chriseth]

I think we need to change the interaction between the type inference and the side effect checker a little.

Because now, statements without assignment are valid everywhere, so we cannot really make this an error in the side effect checker in non-constr functions any more (so the "Tried to add a constraint in a {} context: {expr}",-error needs to go without direct replacement).

Then I think we need to change statement_type to something like constr_function_statement_type (and keep it at the type passed in from outside).

Now whenever we enter the body of a LambdaExpression in type inference, we need to set a flag depending on its kind: When it is constr, we allow both the empty tuple and constr_function_statement_type. When it is anything else, we only allow the empty tuple. At the very outside of type inference, we start out as if it was a constr function.