[llvm] AMDGPU: Use pattern to select instruction for intrinsic llvm.fptrunc.round (PR #105761)

[Thorsten Schütt via llvm-commits]

tschuett wrote:

Many llvm intrinsics that have been raised to named opcodes are in this list, e.g. vscale. 

https://github.com/llvm/llvm-project/pull/105761