intel/compiler/mesh: optimize payload and offset computations
Minor optimizations that reduce the number of instructions used. In some extreme cases (like func.mesh.ext.outputs.per_vertex.indirect_array crucible test), payload computation optimization unlocks some other optimizations and reduces the number of instructions by a factor of 2! No noticeable improvement to vk_meshlet_cadscene performance though.
Depends on !21007 (merged).