[llvm] [AMDGPU] Account for inline asm size in inst_pref_size calculation (PR #192133)

[Adel Ejjeh via llvm-commits]

adelejjeh wrote:

@rampitec I was considering taking the route of actually assembling the inline asm string using MC to calculate accurate length, but I figured under-estimating is safe enough for the prefetch hint and the cost of running each inline assembly instruction through the MCAsmParser might be too high for compile-time. Curious to hear your thoughts!

https://github.com/llvm/llvm-project/pull/192133