[SPIR-V] sign() codegen uses incorrect vector size for non-square uint matrices with at least two rows/columns

Description

sign() generates invalid SPIR-V for any non-square uintMxN matrix with at least two rows and two columns due to a shape mismatch in the generated code: it uses OpUGreaterThan to compare each N-element row vector to an M-element zero vector and uses an M-element boolean vector result type.

Steps to Reproduce

Godbolt

Compile with dxc -T cs_6_0 -E main -Od -spirv source.hlsl:

[numthreads(1,1,1)]
void main() {
  uint2x4 u;
  int2x4 result = sign(u);
}

Actual Behavior

fatal error: generated SPIR-V is invalid: Expected vector sizes of Result Type and the operands to be equal: UGreaterThan
  %31 = OpUGreaterThan %v2bool %28 %8

Disabling validation, we can see the shape mismatch:

          %8 = OpConstantComposite %v2uint %uint_0 %uint_0
         [...]
         %28 = OpCompositeExtract %v4uint %27 0
         %31 = OpUGreaterThan %v2bool %28 %8

Environment

• DXC version: libdxcompiler.so: 1.10(5474-942eeceb)(1.9.0.15470)
• Host Operating System: Fedora Linux 43

Reproduce the issue with the provided uint2x4 HLSL example and the dxc -T cs_6_0 -E main -Od -spirv source.hlsl command, or use the linked Godbolt case. Trace sign() code generation for non-square uint matrices and verify that the generated OpUGreaterThan operands and result type use matching vector sizes and pass SPIR-V validation.