[Linalg] Validation of MatVecMul incorrectly reports wrong vector length

[llvm-beanz]

Description
When constructing a MatVecMul with an FP8 matrix, and FP16 vector inputs and outputs, the validator incorrectly identifies the vector length as invalid. It seems to be assuming that the vector length must be packed disregarding the input interpretation type.

Steps to Reproduce
https://hlsl.godbolt.org/z/zM9zh3ah1

Environment

• DXC version: main
• Host Operating System: all