A void value is treated as linear: using it twice is a move error

void is spelled Type::Struct("void") (syntax/types.rs, is_void_ty), and the linearity rules consume any Type::Struct. So a valueless result behaves like an owned resource:

fn nothing() -> void { }
fn main() -> i32 {
  let z = nothing();
  let a = z;
  let b = z;
  return 0;
}

Error[E4001] at 5:11: Use of moved or consumed linear variable: z

There is nothing to move. The same rule makes a unit-valued expression (an if with no else, an empty block) report a spurious move when used twice.

Found while splitting the tensor spellings (#399). Before that work these sites answered a dims-less Tensor<f32>, which is also linear, so the behaviour predates the split and is unchanged by it — the switch to void only made the diagnostic name the type honestly instead of printing Tensor(F32, [], None) at the user.

Fix direction: exempt void/none from consumption wherever Type::Struct triggers it, or give unit its own Type variant so it cannot inherit struct rules by spelling.

Start in syntax/types.rs with Type::Struct("void") and is_void_ty, then trace the linearity path that consumes Type::Struct values. Reproduce the issue with the provided nothing() example and the unit-valued expression cases. Done means void and unit values can be used more than once without a spurious move error, while other linear structs retain their behavior.