Should all expressions have known types? #723

[edit:] This was previously raised in detail by @kvark here.

In particular, this would preclude certain syntax proposals, impacting the following discussions:

• Syntax for "the zero value" for bool type, numeric types, and compositions of them #685: can't have a syntax like const x: vec3<f32> = {};, can have a syntax like const x = vec3<f32>();.
• Numeric literals and [wgsl] Numeric sizes #575: can't have an "un-narrowed integer literal" type; can't have const x: u32 = 1;, would use const x = 1u;.
• [wgsl] Implicit types for variable declarations with expressions. #560: This makes it always possible to omit type annotations on const/var - they become simple type assertions, as there’s only one valid type annotation. It's even possible to remove them from the grammar (@jdashg proposed this before).

I think this is a desirable trait for simplicity and explicitness of the language, and aside from numeric literals I think it has very little negative impact on authorability.