[llvm] [InstCombine] Improve select simplification based on known bits (PR #97289)

[Nikita Popov via llvm-commits]

nikic wrote:

> Can we revert #66740 after landing this patch? @cyyself found that #66740 caused a significant compile-time regression in SLPVectorizer when building RTL simulator for https://github.com/OpenXiangShan/XiangShan.

I don't really get how the optimization in this PR would help the motivating issues for #66740?

https://github.com/llvm/llvm-project/pull/97289