fp16 uniform mat3 lowering is broken badly somewhere
Zink and llvmpipe have recently come across fails in the GTF-GL46.gtf21.GL.mat3.mat3arraysimple_frag test.
zink wanted to just not allow lowering on desktop GL (leaving this broken for GLES none the less) llvmpipe I was going to disable the fp16 constant buffer cap.
Turns out fp16 mat3 support is broken, and I'm not sure where or how it should be fixed.
mat3[2]: 1.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 1.0 1.0 0.0 0.0 0.25 0.25 0.0 0.0 0.0 0.25
What happens is the uniform uploader puts things like the matrix rows are f16vec4 aligned.
However the io lowering and glsl types don't do the same calcs and we end up reading the wrong values.
I'm not sure where to fix this in the uniform upload or the io (I suspect both need fixing), but we might need to agree on what the const buffer format should actually look like here for packed f16mat3.