[AMDGPU] Refactor AMDGPUTargetStreamer::EmitCodeEnd

Refactor and add comments to explain where the magic numbers come from
in terms of the instruction cache line size. NFC.

Differential Revision: https://reviews.llvm.org/D98266