[llvm] [RISCV] Enable select optimization by default (PR #178394)

[Luke Lau via llvm-commits]

lukel97 wrote:

Not from using this PR, but here are the results for the spacemit-x60: https://lnt.lukelau.me/db_default/v4/nts/1457?compare_to=1455

I think most of the changes are noise from code layout changes etc., the dynamic instruction count is mostly the same.

AArch64 enables this for their default generic tuning info, which seems to be based around the CortexA510Model. 

Won't this help in-order cores too? Should we add it to GenericTuneFeatures

https://github.com/llvm/llvm-project/pull/178394