[llvm] [AArch64][ARM] Optimize more `tbl`/`tbx` calls into `shufflevector` (PR #169748)

[David Green via llvm-commits]

davemgreen wrote:

> I've just rebased this on top of main, without rearranging the InstCombine stuff. I'm not really sure how to move that into its own module in a way that works with `BUILD_SHARED_LIBS`.

This doesn't look any worse than what was already present :)

https://github.com/llvm/llvm-project/pull/169748