[glsl-in] Matrix construction inconsistencies and errors

[Vipitis]

Description
The GLSL Documentation Wiki for Matrix construction isn't that clear to me: https://www.khronos.org/opengl/wiki/Data_Type_(GLSL)#Matrix_constructors and the spec goes into further detail: https://registry.khronos.org/OpenGL/specs/gl/GLSLangSpec.4.50.pdf (chapter 5.4.2, pp. 98..101)

The observation is as follows:

// identity matrix works
mat2 a = mat2(0.1);
    
// from vec2 both work
mat2 b1 = mat2(vec2(0.2));
mat2 b2 = mat2(vec2(0.2), vec2(0.2));

// with floats works
mat2 c = mat2(0.3, 0.3, 0.3, 0.3);
    
//with a vec4 - does not work!
mat2 d = mat2(vec4(0.4));

// some mixture of vec2 and float: e1 and e2 aren't enough components, e3-e5 work, e6 does not
// mat2 e1 = mat2(vec2(0.5), 0.5);
// mat2 e2 = mat2(0.5, vec2(0.5));
mat2 e3 = mat2(vec2(0.5), 0.5, 0.5);
mat2 e4 = mat2(0.5, vec2(0.5), 0.5);
mat2 e5 = mat2(0.5, 0.5, vec2(0.5));
mat2 e6 = mat2(0.5, 0.5, 0.5, vec2(0.5));

Constructing mat2 d throws an error: Composing 0's component type is not expected.
Constructing mat2 e6 also throws and error: Composing expects 2 components but 1 were given (somewhat misleading error)

My reference is Shadertoy where b1 does not work but d works (and e6 as well). Example: https://www.shadertoy.com/view/4cl3Wn
I have no tried matrices of other shapes.
Since a lot of shaders on Shadertoy use mat2 construction in combination with swizzling, this seemingly account for 10% of validation errors pygfx/shadertoy#15.

Platform
using naga-cli 0.14.0 and wgpu-native 0.18.1.3 (all recent glsl commits are missed, it might be addressed already)